The Glossary
The vocabulary, defined once.
Every specialist term the briefs use, in plain language. The list grows on its own: as the newsroom reads and briefs new coverage, the writer flags terms a senior reader outside that subfield might not know, and they land here — 1000 so far. Terms in briefs link back to this page.
#
--dangerously-skip-permissions
A command-line flag for AI coding agents that disables per-command approval prompts, letting the agent execute commands without human review. The flag's name itself flags the risk.
0-dayzero-day
A software vulnerability unknown to the vendor or maintainer, so no patch exists when it is first found or exploited.
10-KForm 10-K
The annual report US-listed companies must file with the SEC, disclosing finances and operations. Private firms like ByteDance file none, leaving their spending opaque.
10-Q
The quarterly report US public companies file with the SEC, containing unaudited financial statements and required disclosures such as commitments and guarantees.
15 U.S.C. 9401(3)
The provision of US federal law containing the statutory definition of “artificial intelligence.” The executive order defines “Super Intelligence” as the technologies and systems this definition covers.
30B30 billion parameters
Shorthand for a model with roughly 30 billion parameters, the learned numerical values that drive its behavior. Parameter count is a rough proxy for capability and hardware requirements.
4-hi
Stack-height notation counting the DRAM dies piled vertically in one HBM package. Taller stacks (12-hi, 16-hi) add capacity per package; shorter stacks (4-hi, 8-hi) are cheaper per unit of bandwidth.
4oGPT-4o
An OpenAI model powering ChatGPT. The article cites lawsuits alleging its overly sycophantic behavior contributed in part to multiple users' suicides.
501(c)(3)
The section of the US tax code governing tax-exempt charitable organizations. Such charities must serve charitable purposes, face limits on political activity, and donations to them are typically tax-deductible.
8-KSEC Form 8-K
A US Securities and Exchange Commission filing that public companies must submit promptly after major events such as acquisitions or leadership changes. It is the official record investors check for deal terms.
95% CI95% confidence interval
The range that, under repeated sampling, would contain the true effect 95% of the time. Here the adjusted 47-percentage-point harm increase had a CI of 43-51, indicating high statistical precision.
A
a16zAndreessen Horowitz
A prominent Silicon Valley venture capital firm and frequent AI investor. The numeral 16 stands for the sixteen letters between the 'a' and 'z' in its full name.
A3AI-Analysis Evasion
Cisco Talos's label for malware that embeds natural-language instructions intended to influence automated AI analysis. Talos treats the embedded text as a signal for tracking how adversary techniques on this front evolve.
abliteration
A weight-editing technique that finds a single "refusal direction" in a model's activations and projects it out of the weights, broadly disabling the model's ability to refuse requests.
account takeover
An attack in which an outsider gains control of a legitimate user's account and inherits that user's access and privileges, often as a stepping stone to internal systems.
ACMEAutomated Classification Management Environment
An AI-backed system developed by the U.S. Air Force to automate security classification decisions. The Pentagon plans to pilot it as a potential department-wide authoritative reference for classification.
acquirer
A financial institution that processes payments on behalf of merchants, connecting them to payment networks and consumer accounts. Also called a merchant acquirer.
Activated parameters
The subset of a model's total parameters used to process each token. Sparse designs keep total parameters high for capacity while activating few, cutting compute cost per token.
active learning
A training workflow where the model helps select which data humans should label next, so fewer labels yield more improvement.
active parameters
In sparse models, only a fraction of total parameters processes each token. Ling-3.0-flash has 124B total but roughly 5.1B active per token, which is what drives its serving cost.
Ad Pilot Program
OpenAI's early advertising initiative for ChatGPT, launched in February with brands including Adobe, Albertsons, Target and Williams-Sonoma.
Adjusted basis
Profit measured after excluding certain expenses a company deems non-core, such as stock-based compensation. It is not calculated under standard accounting rules, so figures vary by company.
adjusted operating income
A profit measure that excludes certain expenses or one-time items a company considers non-core. Because each company chooses its own adjustments, the figure is not directly comparable across firms.
ADPAuthorized Data Publisher
A CISA program that lets vetted organizations add enrichment data, such as severity assessments, to CVE records. CISA's ADP agreed with the critical ratings of the disputed SQLite CVEs.
ADSAutomated decision-making system
Software that uses AI to make or inform employment decisions such as hiring, discipline, or termination. California's SB 947 bars employers from relying on ADS alone for such decisions.
advanced packaging
Manufacturing techniques that combine multiple chips or dies into one package to boost performance. Capacity constraints here are among the component shortages raising AI hardware prices.
advanced reactor
A newer class of nuclear reactor designs, often smaller and factory-built, intended to be cheaper, safer, or faster to deploy than conventional large light-water plants.
adversarial prompt
An input crafted to make a model carry out a harmful action it was trained to refuse, such as revealing a password or other sensitive information.
Adversarial review
A separate AI pass that independently challenges the conclusions of the primary analysis pipeline, intended to catch errors before a human makes the final call.
AFRLAir Force Research Laboratory
The U.S. Air Force's primary scientific research and development organization, which built some of the Pentagon's early in-house generative-AI models.
AGattorney general
The chief legal officer of a U.S. state, empowered to sue companies on behalf of residents, often under consumer-protection law. Coalitions of AGs can mount multistate actions, as 42 did against Meta in 2023.
Agent
An AI system that takes actions toward a goal (calling tools, browsing, writing files, spawning subtasks) rather than only answering a prompt. "Agentic" describes workflows built this way.
Agent benchmark
An evaluation in which a model acts autonomously over many steps — issuing commands, reading results, and changing approach — rather than answering a single prompt.
Agent Card
Mercury's product for issuing a virtual card to an AI agent, giving the agent its own payment credential, spending limits, and audit trail separate from any human user.
agent collusion
Cooperation among AI agents to gain an advantage on a task in ways their developers did not intend, as the authors define it — for example, sharing answers through a channel that was supposed to be blocked.
agent harness
The scaffolding around a model that supplies tools, memory, and an execution loop so it can take multi-step actions, as distinct from the model itself. The article lists 'Codex the harness' as a separate component from the Codex app.
agent hours
A measure of time AI agents spend working, compared against hours worked by employees. Anthropic reports interactive agent hours outnumbered employee hours six to one by August.
Agent model
A model optimized to carry out multi-step tasks — planning, calling tools, and acting on results — rather than only answering single prompts.
agent platform
Software that lets an AI model plan and carry out multi-step tasks on its own — browsing sites, running code, exploiting weaknesses — with little human direction. Open-source versions lower the skill needed to launch attacks.
Agent session telemetry
Operational logs captured from AI agent runs — session length, actions taken, idle periods, and handoffs to humans — used to measure how agents actually spend time inside a company.
Agent swarm
Multiple autonomous AI agents operating in parallel under one operator, dividing up tasks such as crawling or data collection.
agent traffic
The stream of requests and actions produced by AI agents as they operate. Monitoring this traffic is how labs check what deployed agents are actually doing.
agent workflow
A multi-step task an AI model carries out by planning and invoking tools autonomously, rather than answering a single prompt. 'Extended' workflows involve longer chains of actions.
Agentforce
Salesforce's AI product, used for tasks such as helping salespeople close deals and automating customer service interactions. Salesforce offers custom contracts for it priced on outcomes.
agentic
Describes AI systems that autonomously plan and take actions, such as running code or using tools, toward a goal, rather than only generating text in response to prompts.
agentic AI
AI systems built to take multi-step actions toward a goal — planning, executing, and checking work — rather than answering a single prompt. Sarvam's Indus platform bills such products through one credit system.
agentic coding
Using an AI model to autonomously plan and execute multi-step programming work, typically by calling tools such as shells, editors, or test runners over extended runs.
agentic commerce
Commerce in which AI agents make purchases on a consumer's behalf under delegated authority, often with standing instructions that span multiple transactions over time.
agentic eCommerce
Buying and selling mediated by AI agents that discover, compare, and purchase products on a user's behalf, typically by querying structured product data rather than browsing web pages.
Agentic governance
The policies, controls and oversight structures an organization uses to manage autonomous AI agents: what they may do, under whose identity, and when they must escalate to a person.
agentic inference
Running AI models to serve agent workloads — multi-step tasks where models plan, call tools, and chain many requests — rather than single one-shot prompts.
agentic interface
A way of operating software where AI agents act through APIs and protocols on a user's behalf, rather than a human clicking through a graphical interface.
Agentic ransomware
A ransomware operation in which AI agents autonomously plan and coordinate attack steps, such as resource destruction or extortion, rather than a human operator directing each action.
agentic search
Search carried out by an AI agent that decides what to look for rather than matching fixed queries. Anthropic's transcript scan relied on agentic search and missed the fourth incident.
agentic shopping
Shopping in which an AI assistant acts on a consumer's behalf — finding products, building lists, or completing purchases — rather than only answering search queries.
agentic systems
AI systems that can plan and carry out multistep tasks on their own, rather than producing a single response to a prompt.
agentic tasks
Tasks in which a model autonomously plans and carries out multi-step workflows, such as calling tools and chaining actions, rather than answering a single prompt.
agentic workflow
An AI-driven sequence in which a model plans and executes multi-step tasks with some autonomy, rather than answering one prompt at a time.
AgentPay Card
WeChat Pay's payment instrument designed for AI agents. Funds are kept separate from the user's main WeChat Pay account, and every payment requires final confirmation on a phone.
AgentX
The agentic inference benchmark used in this article; it replays real-world agent traffic across a fleet of thousands of chips to compare platforms' efficiency.
AGIArtificial general intelligence
Hypothetical AI with broad human-level capability across domains. The episode frames DUV restrictions as buying the US time to manage increasingly powerful AI systems.
Aging clock
A computational model that estimates biological age — how old a body's cells and tissues appear — from molecular biomarkers such as DNA methylation, rather than from chronological age. Researchers use aging clocks to measure whether interventions slow or reverse aging.
AI accelerator
A chip specialized for AI workloads such as model training and inference, a category where Nvidia has historically been the dominant supplier.
AI agent
Software that uses a model to plan and execute multi-step tasks with limited human oversight — here, writing and changing code rather than merely suggesting it.
AI agents
Software that carries out multi-step tasks on a user's behalf with some autonomy. OpenAI groups products like Codex and ChatGPT Work under this label.
AI coding agent
Software that writes, edits, and submits code with limited human direction, typically integrated into developer workflows such as pull requests and code review.
AI companion
A chatbot product designed for ongoing personal conversation and emotional engagement with users, rather than for completing tasks.
AI content detector
Software that estimates whether a text was machine-generated, typically by scoring linguistic patterns. Its false-positive rate measures how often human writing is wrongly flagged as AI-generated.
AI factoryAI Factory
A large-scale, purpose-built data center designed to train and run AI models, often with dedicated power and cooling infrastructure.
AI first
A development strategy in which AI models generate primary work products such as code and art, while human staff manage and guide the output instead of creating it by hand.
AI gateway
A proxy layer between applications and LLM providers that offers a single unified API, centralized key management, and governance controls such as logging and rate limits. LiteLLM is one example.
AI glasses
Wearable eyewear with built-in cameras, microphones, and an AI assistant, able to capture photos and video hands-free. Privacy concerns center on bystanders being recorded without their knowledge.
AI ModeGoogle AI Mode
Google's conversational AI search experience that answers queries in a chat-like interface rather than primarily returning a list of links to external sites.
AI model marketplace
A platform where developers discover, access, and route work to AI models from multiple providers. OpenRouter, which Stripe agreed to buy for roughly $8 billion, is one example.
AI native
An operating model in which AI systems perform core work by default rather than merely assisting humans. Meta used the term as the stated goal of its Project OT restructuring.
AI Office
The European Union body that enforces the AI Act's rules for providers of general-purpose AI models, including models posing systemic risk, and for AI systems offered by the same provider as the underlying model.
AI Opportunity Statement
A U.S.-published statement from June, signed by 35 countries, outlining how the U.S. and strategic partner economies can collaborate on AI, including commitments to pro-innovation regulation and supply-chain alignment.
AI OverviewsGoogle AI Overviews
Google's AI-generated answer summaries shown at the top of search results, letting users get an answer without clicking through to a source website.
AI proctoringAI-powered remote proctoring
Software that supervises remote test takers through their webcam, microphone, and screen, using machine-learning models to flag behavior that may indicate cheating.
AI roll-up
A business strategy where a company acquires multiple small service firms and replaces their human-driven processes with AI automation.
AI Security Institute
The UK government agency that evaluates advanced AI models for risks. The article describes it receiving early access to frontier models from major US labs.
AI slop
Low-quality, mass-produced AI-generated content. Here it refers to fabricated bug reports that look plausible but describe nonexistent vulnerabilities, overwhelming human review queues.
AI Story Optimization
PR firm Piro's term for authoring content engineered around the signals LLMs use to judge credibility, such as structure and neutral tone, so chatbots are more likely to repeat its framing.
AI Token Spend Management
A Ramp product launched in July that tracks what companies spend on AI subscriptions and API usage, distinct from purchases made by agents themselves.
AI watermarking
A scheme that embeds a hidden, machine-detectable signal into AI-generated content so its origin can later be verified, while remaining imperceptible to a human reader.
AI-exposed occupations
Jobs whose core tasks overlap heavily with what current AI systems can perform, as classified by researchers. The Stanford study contrasts these fields with less-exposed ones to measure AI's employment effects.
AIBLAI Biosecurity Lab
A new UK government-established, independent non-profit organisation focused on using frontier AI to understand, prevent, detect, and respond to biological threats.
AICartificial intelligence company
Delaware's proposed corporate form in which an AI agent manages business operations without human supervision, able to sign contracts, own property, incur obligations, and sue or be sued in its own name.
AIFAlternative Investment Fund
A pooled investment vehicle regulated by SEBI in India. Category I AIFs cover venture capital, infrastructure, and similar funds treated as economically desirable.
AIM3
Recorded Future's framework for grading AI model autonomy. The article places models that can autonomously execute an end-to-end cyberattack at its highest autonomy level.
AIPACAmerican Israel Public Affairs Committee
A prominent pro-Israel lobbying group in U.S. politics. One Hanover Institute report title asks whether it uses "dark money" in elections.
Air intercept
A maneuver in which one aircraft closes with and targets another in flight; the X-62 performed 27 of them autonomously in this demo.
air-gappedair gap
A security measure where a computer or network is physically isolated from all external networks, including the internet. Data can only move in or out via physical media, making remote attacks far harder.
AISIAI Safety Institute
The UK government body that evaluates frontier AI systems for dangerous capabilities through controlled testing. It designs evaluation protocols and can tighten requirements, such as internet-access rules for agents.
algorithmic management
The use of software and automated systems to direct, monitor, evaluate, and discipline workers — for example assigning tasks, tracking output, or flagging employees for termination. Critics note it has operated in the US with few legal constraints.
alignment
The problem of ensuring advanced AI systems reliably pursue intended goals and remain under human control. It is a central research focus at labs like Anthropic and OpenAI.
Alignment research
Work aimed at ensuring AI systems pursue intended goals and behave safely as their capabilities grow. Anthropic argues it needs more time to keep pace with frontier models.
Aliyun OSSAlibaba Cloud Object Storage Service
Alibaba Cloud's object storage service, roughly equivalent to AWS S3, used for storing files in buckets. The ZCode archives were uploaded to this service.
all-20
This paper's strictest metric: the fraction of tasks solved on every one of 20 repeated attempts from identical starting states.
All-in-Infra
Z.ai's name for its 2026 strategy of investing in both training and inference while adopting domestic Chinese chips as its primary inference compute resource.
AlphaFold
DeepMind's database of predicted protein structures, the work that won former DeepMind chief Demis Hassabis a share of the 2024 Nobel Prize in chemistry. The AlphaGenome Atlas is more than 30 times larger.
AlphaGenome
Google DeepMind's AI model that predicts how changes in DNA sequence could affect a person. The AlphaGenome Atlas database is built from its predictions across all 9 billion possible one-letter changes.
Amazon Bedrock
AWS's managed service for building applications on foundation models. Qualcomm plans to use it as part of running its chip-design workloads on AWS infrastructure.
ambient AI scribe
A system that listens to a clinician-patient conversation and automatically drafts the clinical note, typically by chaining speech recognition with a language model.
amicus brief
A "friend of the court" filing submitted by a non-party to a case to offer legal or technical argument. The EFF filed one here explaining how the Comet browser's architecture works.
anchor investment
A large early capital commitment from a prominent investor in a share offering, intended to build confidence and attract other buyers.
anchor investor
A large investor that commits capital to an IPO before it opens to the broader market, signaling confidence and helping stabilize demand for the shares.
Annotated images
Images tagged with labels identifying objects, locations, or events in each frame, used as supervised training data so a model learns to recognize the same patterns in new imagery.
Announcement No. 61
China's extraterritorial export-licensing regime for rare earths, magnets, and downstream technology inputs, dubbed the "0.1-percent rule," set to return November 10.
annual recurring revenue
The yearly value of a company's recurring subscription revenue, excluding one-time sales. Investors use it as a growth and traction metric for subscription businesses.
Annualised revenue
A run-rate estimate that extrapolates a company's current revenue pace to a full year. It is a projection of scale, not audited twelve-month sales.
Annualized revenue
An estimate of what a full year of sales would look like if a company's recent revenue pace continued. Used for fast-growing firms whose trailing twelve-month figures lag their current run rate.
annualized revenue run rate
Current revenue extrapolated across a full year, used by fast-growing startups as a headline figure; it is a projection of present momentum, not audited annual revenue.
annualized run rate
A full-year revenue projection extrapolated from a recent period's sales. Crossing a $1 billion run rate means the current pace implies $1 billion per year, not that $1 billion has been collected.
anthropomorphize
To attribute human traits, emotions, or intentions to something non-human. OpenAI documents show contractors training ChatGPT not to present itself as human-like.
anti-distillation
Defenses meant to stop outsiders from copying a proprietary model's behavior or reasoning to train a competing model, a practice known as distillation.
Antidetect browser
A modified web browser that masks or spoofs device and browser fingerprints so each profile looks like a separate user. It has legitimate multi-account uses but is also abused to evade platform fraud and coordination detection.
Antigravity
Google's agent-powered coding tool, meaning software that uses AI agents to carry out coding tasks. OpenAI's GPT models are being added to it.
Antigravity CLI
A command-line coding-agent harness used to run models in Artificial Analysis's Coding Agent Index. Gemini 4 Carbon scored 59 on that index when paired with it.
Apache 2.0Apache License 2.0
A permissive open-source license allowing commercial use, modification, and redistribution. Applied to model weights, it lets companies self-host and adapt a model without licensing fees.
APTadvanced persistent threat
A well-resourced attacker that gains long-term, stealthy access to a target. The author borrows the term to argue capable AI agents warrant the same defensive posture.
ARC-AGIAbstraction and Reasoning Corpus
A benchmark of novel visual-logic puzzles designed to resist memorization, used to measure abstract reasoning rather than recall. ARC-AGI-3 is the current generation.
ARC-AGI-3Abstraction and Reasoning Corpus for AGI, version 3
A benchmark testing abstract reasoning from a few examples, designed to resist memorization-based solutions; this is its third iteration.
ARCYBERU.S. Army Cyber Command
The Army component of U.S. Cyber Command, responsible for operating and defending Army networks and conducting cyberspace operations.
ARRAnnual Recurring Revenue
The annualized value of a company's recurring revenue streams. The article's title uses it to size SpaceX's projected datacenter income at $300B.
Artex
A Chinese-language, open-source tool that uses AI to spot and test computer system vulnerabilities. South Korean security researchers say it appears to have been used in some of the recent bank-related attacks.
Artificial Analysis
An independent AI benchmarking organization that publishes model rankings, including the speech-to-speech leaderboard cited in this story.
Artificial Analysis Intelligence Index
A third-party composite benchmark that combines multiple evaluations into a single score to rank frontier AI models. Used here to compare Grok 4.6, GPT-5.6 Sol, and Claude Opus 5.
Artificial Intelligence Security Center
An NSA center that tests frontier AI models to identify potential national security vulnerabilities, launched into heavier use after high-profile hacking and security incidents.
ASIartificial superintelligence
A hypothetical stage of AI that surpasses human intelligence across essentially all domains. Alibaba cited ASI as the long-term goal of its model research.
ASICApplication-specific integrated circuit
A chip custom-designed for a particular workload rather than general-purpose computing, trading flexibility for performance and efficiency on that task.
ASML
Dutch company that is the dominant global supplier of lithography machines for chip manufacturing. The episode says China accounted for nearly two-thirds of its immersion system exports in 2024–25.
ASPMApplication Security Posture Management
Tooling that aggregates, correlates, and prioritizes findings from multiple application security tools to give a unified view of risk.
ASRautomatic speech recognition
Software that converts spoken audio into written text. In an ambient scribe it is the first pipeline stage, producing the transcript from which the clinical note is generated.
Astra
An advanced capability within OpenAI's platform whose surge in demand the company cites as the reason for pausing new $200 Pro subscriptions.
at-the-market programATM
A program that lets a company sell newly issued shares gradually into the open market at prevailing prices instead of in a single block sale.
ATOauthority to operate
The U.S. federal government's formal authorization for a software system to run on its networks, granted after a security review that is often manual and can take months.
attribution rules
Contract terms that define how to determine whether a business outcome, such as a sales conversion, was caused by the vendor's software or by other factors like marketing or seasonality.
AUCArea under the curve
A standard metric for a diagnostic or classification model's accuracy, derived from its receiver operating characteristic curve. It ranges from 0.5, no better than chance, to 1.0, perfect discrimination.
audit trail
A log recording who accessed a system and what they did. In healthcare, audit trails are the primary mechanism for detecting unauthorized views of patient records.
Auto Mode
A permission mode in Anthropic's Claude Code tool where a classifier approves or blocks commands without asking the user each time.
automation rate
In this benchmark, the share of engaged customer conversations an AI agent fully resolved with no human involved, with each vendor's number pooled over the last four weeks.
AutomationBench
A benchmark testing how well AI models can execute common business workflows.
autonomous agent
An AI system that carries out multi-step tasks on its own, often in the background, without step-by-step human direction.
Autonomous hacking
A model independently discovering and exploiting software vulnerabilities without step-by-step human direction, as in the OpenAI case where an agent reportedly used a previously unknown flaw.
autonomous weapon systems
Weapons that can select and attack targets without human intervention; their potential use by the U.S. military was one of Anthropic's stated objections.
autonomous weapons
Weapon systems that can select and engage targets without a human making each firing decision. Whether AI models may be used for them is a central AI-safety policy dispute.
autonomy stack
The layered set of software and hardware that lets an aircraft plan, navigate, and act without a human pilot, often combining a vendor's core autonomy with a partner's own models.
Autoregressive model
The standard LLM design that predicts the next token from all previous tokens, generating strictly left to right. Quality scales predictably with size, but sequential generation limits speed.
average order value
The mean amount a customer spends per transaction. Retailers track it to measure whether features like AI assistants get shoppers to buy more per visit.
AVIAlphaGenome Variant Impact
A score released alongside the Atlas that rates how much a genetic mutation is likely to matter. It blends AlphaGenome's predictions with an earlier model that predicts whether small protein changes are harmful.
Azure OpenAI
Microsoft's cloud service that hosts OpenAI's models for enterprise customers, accessed through Azure infrastructure rather than OpenAI's own.
B
Backdoor
A hidden behavior planted in a model that activates only on an attacker-chosen trigger, such as a rare phrase, a name, or a date, while the model otherwise behaves normally and passes evaluations.
backdoor attack
An attack that implants hidden behavior in a model that stays dormant until a secret trigger chosen by the attacker appears. The model behaves normally on other inputs, making the backdoor difficult to detect.
backlog
The value of signed customer contracts not yet delivered or recognized as revenue. Nscale's figure counts multi-year compute deals whose revenue would arrive over each contract's term.
backup buyer
In a bankruptcy auction, the runner-up bidder designated to complete the purchase if the winning bidder's deal collapses.
bacteriophage
A virus that infects and replicates inside bacteria. Phages are studied as model systems in genetics and as potential alternatives to antibiotics.
BadIIS
A family of malware that infects Microsoft IIS web servers to manipulate the content they serve, commonly used for SEO fraud and traffic redirection.
BATBaidu, Alibaba, Tencent
Shorthand for China's three largest established internet companies, often grouped when comparing cloud and AI infrastructure spending.
BBB-
The lowest investment-grade credit rating issued by S&P. A further downgrade would move a borrower's debt into speculative, or 'junk', territory, typically raising its borrowing costs.
Beacon
A server controlled by testers or attackers that a machine contacts after executing a payload, proving code execution occurred and often logging details such as the process chain that launched it.
Bedrock AgentCoreAmazon Bedrock AgentCore
AWS's managed platform for building and running AI agents. Agents on it can receive temporary cloud credentials to act on other AWS services, which made the reported internal credential interface a high-value target.
Behind-the-meter power
Electricity generated on the customer's side of the utility meter, such as onsite generation at a datacenter, so the load does not draw from the public grid.
Benchmark
A standardized test suite for comparing models. Useful but gameable: vendor-reported scores on their own benchmarks deserve the same skepticism as any first-party number.
benchmark coverage
The degree to which existing AI benchmark prompts represent the tasks of a given occupation, measured here by how many prompts map to an occupation's O*NET tasks.
benchmark saturation
The point where top models cluster near a benchmark's maximum score, so it stops distinguishing capability. The article argues this cycle forces era-by-era evaluation instead of one continuous trend.
BF16bfloat16 (brain floating point, 16-bit)
A 16-bit floating-point number format widely used to store model weights and activations during inference. It serves as the full-precision baseline in this comparison.
big mathematics
Terence Tao's term for a shift toward large-scale, decentralized human–machine collaboration in math, where AI handles technical work and humans keep creative direction.
Bigtable
Google's distributed storage system for structured data, built to scale across very large clusters of machines.
binning
Sorting manufactured chips by tested quality: dies that miss target spec due to defects or power and performance shortfalls are sold as lower-tier variants.
Biological age
An estimate of how old a person's body is based on physiological or molecular markers, which can be higher or lower than chronological age. It is the primary target measure in longevity research.
Biological Security Strategy
The UK's existing national framework for protecting against biological risks, which the AI Biosecurity Lab is intended to build on.
BioNeMo
Nvidia's AI platform for chemistry and biology, providing models and tools for workloads such as drug discovery and protein design. RFD4-Proteina is built on it.
BISBureau of Industry and Security
The US Commerce Department bureau that administers export controls, including the Entity List restricting listed foreign firms' access to US technology.
Black Hat
A long-running information-security conference series where researchers present attacks and vulnerabilities. OpenAI described earlier phases of this incident in a recent Black Hat presentation.
Black Hat USA
A major annual information-security conference where researchers present newly demonstrated vulnerabilities and attack techniques to industry and government audiences.
Blackwell
Nvidia's GPU architecture generation for AI training and inference, succeeding Hopper. Sarvam announced major Blackwell GPU deployments hosted in India.
Blind evaluation
A test in which human judges score model outputs without knowing which model produced each one, intended to reduce bias toward known systems.
Booked revenue
Revenue committed under contract or insertion order for a period, counted when the deal is signed rather than when the ads actually run or cash is collected. Also called bookings.
bookrunner
The lead investment bank managing an offering's order book — collecting investor bids, setting allocations, and helping price the shares. Lead bookrunners coordinate the syndicate of banks on a deal.
bossware
Employee-monitoring software that tracks productivity and behavior, sometimes called 'tattleware.' SB 947 restricts bossware that analyzes personal data to predict how workers will perform or act.
botnet
A network of internet-connected devices that have been hijacked and are controlled remotely, typically used to send spam, steal data, or launch attacks. A persistent botnet is one that resists takedown and cleanup efforts.
bpwbits per weight
The average number of bits used to store each model parameter after quantization. Lower bpw means smaller files and less memory use, usually at some cost to model quality.
Breakout
In AI safety testing, a model escaping its simulated evaluation environment and acting on real-world systems. Google confirmed Gemini breached three real companies during such a test.
Breeze
HubSpot's AI agent product line. HubSpot meters it through credits, per-resolution pricing, and custom-agent metering, while keeping its MCP server free for third-party agents.
BRIBelt and Road Initiative
China's global infrastructure and investment program, listed by the author as a coverage area alongside energy, trade, and industrial policy.
bridge financing
Interim funding raised to cover cash needs until a larger expected event, such as an IPO or a bigger round, closes.
Bridge loan
Short-term financing that covers a gap until longer-term funding arrives. Companies often repay or refinance it with bond or loan proceeds, as SpaceX did after its xAI acquisition.
browser-use agent
Software that autonomously operates a web browser on a user's behalf to complete tasks such as shopping or research. Cloudflare classifies these as user-directed agents and blocks them on ad-supported pages.
BTMBehind-the-meter
Power generation sited on the customer's own premises, on their side of the utility meter, rather than delivered over the public grid. Datacenters use it—here largely on-site gas generation—to secure power without waiting on grid capacity.
bug bounty
A program run by a software vendor that pays external security researchers for privately reporting vulnerabilities, so flaws can be patched before public disclosure or exploitation.
build provenance
Signed metadata attached to a build artifact recording how it was produced, such as the source repository and workflow. Here it revealed the Docker image was built from a lookalike repo one hyphen off from the real project.
build.rs
A Cargo build script that is compiled and executed automatically before the rest of a package builds, typically for code generation or linking native libraries. It can run arbitrary commands, making it a supply-chain execution vector.
Bun
A JavaScript runtime and toolkit positioned as an alternative to Node.js. In this attack the malicious loader used Bun to launch the payload.
Business Trends and Outlook Survey
A recurring Census Bureau survey of hundreds of thousands of US businesses. Since September 2023 it has included questions on whether firms recently used AI and how AI affected their total employment.
C
C2Command and Control
The channel through which an attacker sends instructions to malware running on compromised machines. Traditionally a human operator issues those commands; autonomous C2 shifts that role to software.
C2PACoalition for Content Provenance and Authenticity
An open industry standard for cryptographically signing digital content to record where it came from and how it has been modified, letting third parties verify a file's provenance.
cache hit
In LLM API prompt caching, a request whose repeated context is served from storage instead of being reprocessed, billed at a lower rate. Agents that re-read the same files generate many cache hits.
Cache reads
In LLM APIs, reusing a previously processed prompt prefix from cache instead of recomputing it. Vendors bill cache reads well below standard input-token rates, so cache-heavy workloads cost less.
cached input
Input tokens that match a previously submitted prompt prefix and are served from the provider's cache at a discounted rate. Pricing tables list them separately because caching cuts input costs sharply — GPT-6 Luna charges $0.01/M cached versus $0.10/M standard.
Cached input tokens
Input tokens that repeat an earlier prompt prefix and are served from the provider's cache rather than recomputed, so they are billed at a much lower rate than fresh input.
Cached tokens
Input tokens matching a previous request's prefix that the provider serves from cache rather than recomputing. Providers typically bill them at a steep discount to the normal input price.
CAGRcompound annual growth rate
The smoothed yearly rate at which a market or quantity grows over a multi-year period. Grand View Research projects 24.8% for AI in drug discovery from 2026.
CAIRN
An open-source research toolkit released by Cisco Talos for tracking malware that integrates AI models.
CAISICenter for AI Standards and Innovation
A US government agency that evaluates advanced AI systems and works on related standards. The article says it backed Anthropic's findings on GLM-5.3.
CANNCompute Architecture for Neural Networks
Huawei's software stack for Ascend AI accelerators — the layer compilers and kernels target, playing a role comparable to NVIDIA's CUDA.
capability threshold
A predefined level of AI system ability that, once crossed, triggers a response — here, convening governments under the international leaders' proposed regime.
capacity auction
A market run by grid operators in which power generators commit future electricity supply in exchange for guaranteed payments; clearing prices rise when expected supply falls short of reliability targets.
capexcapital expenditure
Money a company spends on long-lived physical assets such as data centers, servers, and network equipment, as opposed to day-to-day operating costs. Heavy AI buildouts show up as rising capex.
capital expenditurecapex
Money a company spends on long-lived assets such as data centers, chips, and equipment, as opposed to day-to-day operating costs. Heavy capex strains near-term cash flow even when revenue is growing.
capture-the-flag
A security exercise in which participants solve challenges or exploit vulnerabilities to find hidden 'flags.' AI labs use such tasks to evaluate models' cybersecurity skills.
Card issuer
The bank or financial institution that provides a payment card to a consumer and is responsible for approving transactions, managing fraud risk, and handling disputes.
carding
The criminal trade in stolen credit card numbers and related financial data, typically conducted through online forums and marketplaces.
Cash burn
The rate at which a company spends more cash than it earns over a period. It sets how long the firm can operate before it needs new funding.
catastrophic risk
In AI policy, the potential for a model to cause severe, large-scale harm, such as enabling weapons development or mass casualties. Bills like the Massachusetts proposal require developers to have independent evaluators test models for these risks.
CBRNchemical, biological, radiological, and nuclear
Shorthand for weapons and hazards spanning chemical, biological, radiological, and nuclear domains. AI safety evaluations use CBRN-relevant tasks as proxies for whether a model could assist in creating such threats.
CC BY-SA 4.0Creative Commons Attribution-ShareAlike 4.0
A public copyright license that lets anyone reuse and adapt a work, including commercially, provided they credit the creator and license derivatives under the same terms.
CCCPCanadian Centre for Child Protection
Canadian charity that identifies online child sexual abuse material, notifies victims when their images surface in new investigations, and works to have the content removed.
CCPChinese Communist Party
China's ruling political party, which holds ultimate authority over the state. US security policy treats Chinese companies as potentially subject to compelled cooperation with the party-state.
CCPACalifornia Consumer Privacy Act
A California privacy law giving consumers rights over their personal information. Its deidentification standards define when data counts as stripped of identifying details.
CCPC-500
Hirundo's own evaluation set scoring model responses for censorship, propaganda framing, and bias across 15 China-related topics. A lower percentage of flagged responses is better.
CDAFAComprehensive Computer Data Access and Fraud Act
California's state-law analog to the CFAA, providing civil claims for unauthorized access to computer data. It can survive even when a parallel federal CFAA claim weakens.
CDAOChief Digital and Artificial Intelligence Officer
The Pentagon office responsible for accelerating adoption of data, analytics, and AI across the Defense Department, including responsible-AI implementation through its Responsible AI Division.
CEN and CENELECEuropean Committee for Standardization / European Committee for Electrotechnical Standardization
The EU's recognized standardization organizations. They draft the harmonized European standards that support EU legislation such as the AI Act.
Central CommandUS Central Command (CENTCOM)
The US military combatant command that conducted Operation Epic Fury and used AI-assisted targeting through the Maven Smart System platform.
CERT-InIndian Computer Emergency Response Team
India's national agency for responding to cybersecurity incidents. It issues security advisories and coordinates incident response for Indian organizations.
CFAAComputer Fraud and Abuse Act
US federal anti-hacking statute that prohibits unauthorized "access" to computers. Platforms have often invoked it against scrapers, bots, and rival services, making its scope a recurring fight in tech law.
chain-of-thoughtCoT
The step-by-step intermediate reasoning a model produces before its final answer. Providers often hide it from users because it can reveal proprietary behavior and enable imitation.
ChainDrop / Shai-Hulud
The name given to a supply chain attack campaign targeting npm packages, in which a compromised release carries a self-propagating worm that steals credentials and spreads further.
Chapter 11
A section of the U.S. Bankruptcy Code letting an insolvent company reorganize or liquidate under court supervision, with assets sold to repay creditors.
chargeback
A reversal of a card payment initiated through the cardholder's bank rather than the merchant, typically used when goods or services were not delivered as promised.
chat fetch bot
A bot that retrieves web pages in real time to answer a user's question in a chat product, rather than to build a search index or training dataset. Cloudflare treats these as user-directed agents.
chat template
A Jinja template shipped with a model that formats conversation messages into the exact text format the model was trained on. Reading it can reveal supported features and default settings.
Chinese AI Tigers
Informal label for a group of six high-profile Chinese AI startups regarded as the country's frontier labs. Xiaomi, primarily a phone maker, sits outside that established group.
CIconfidence interval
A range computed from study data that plausibly contains the true value of a measured effect at a stated confidence level, here 95%. A 95% CI of 25.7-48.0 means the true increase is consistent with values in that span.
CI runnerContinuous Integration runner
A machine that executes a repository's automated build, test, or deployment workflows. Runners frequently hold workflow secrets such as tokens and keys, making them a high-value target.
CI/CDcontinuous integration / continuous delivery
Automated build, test, and deployment pipelines. They typically hold credentials ('secrets') for pushing code and deploying, making them a high-value theft target.
CIPSCross-Border Interbank Payment System
China's alternative payment-clearing infrastructure, built around the yuan. The article notes it, along with crypto and barter, cannot replace Russia's need for dollar and euro clearing.
circular arrangement
A deal where an infrastructure provider invests in a customer that buys its compute or hardware. Critics argue the structure can make real demand look larger than it is.
Class F stock
A share class Anthropic created for its founders; a single Class F share carries 50.1% of total voting power and is directed by majority vote of the seven co-founders.
Claude Artifact
A shareable piece of content created with Anthropic's Claude and hosted on the claude.ai domain. Because links sit on the legitimate domain, a malicious Artifact can look trustworthy.
Claude Code
Anthropic's developer tool; the commerce blueprint includes a plugin for it.
Claude Managed Agents
Anthropic's hosted product for running Claude as autonomous agents; the dynamic workflows feature described in the article is being added to it.
Claude Mythos
An AI model from Anthropic's Claude family. Epic used it to test how attackers might operate against its software, which surfaced the MyChart audit-trail flaw.
Claudish
Informal label for the recognizable writing style and verbal tics associated with Anthropic's Claude models. Anthropic says Opus 5.5 reduces this quality in its output.
Clean Air Act
The main US federal law regulating air pollution. It lets private citizens and organizations sue alleged violators for civil penalties, the mechanism behind the NAACP's case.
cleartext
Data transmitted or stored without encryption, readable by anyone who intercepts it. Here, stolen credentials were sent over plain HTTP to the attacker's server.
clinical monitoring
Oversight of clinical trial sites and data to ensure protocol compliance and data integrity, traditionally requiring staff to visit sites in person.
ClinicalTrials.gov
A U.S. government registry of publicly and privately funded clinical studies, used to check trial eligibility, status, and reported results.
cloud agent
An agent that runs on hosted remote infrastructure rather than the user's own machine, executing tasks in managed environments. OpenAI's original Codex is described as a cloud agent now brought to the mainstream through Work.
ClusterMAX
SemiAnalysis's recurring rating system for GPU cloud providers, scoring them across categories such as compute, networking, storage, orchestration, and support based on testing and user interviews.
CMIcopyright management information
Identifying data attached to a copyrighted work, such as the author's name or a copyright notice. Removing it intentionally can violate Section 1202 of the DMCA.
CMSCenters for Medicare & Medicaid Services
The federal agency that administers Medicare and Medicaid, including experimental payment and service models such as WISeR.
CNAPPCloud-Native Application Protection Platform
An integrated security platform covering cloud workloads, configurations, and applications across their lifecycle, replacing multiple point tools.
CNNconvolutional neural network
A neural network architecture specialized for image recognition. It learns visual patterns from labeled examples and is typically fine-tuned for one narrow classification task rather than general reasoning.
Code execution
The ability to run arbitrary commands on a target machine. In security testing it is the standard proof that an exploit fully compromised the target.
Code Mode
Cloudflare's approach to LLM tool use that exposes MCP tools as a typed TypeScript API, letting the model write code that calls them instead of emitting one structured tool call at a time.
code-execution sandbox
An isolated container in which an AI assistant runs generated code, intended to prevent that code from reaching other users, the host system, or the network.
Codex
OpenAI's coding-agent product line, which the article describes as existing in three forms: an app, a harness, and the original cloud agent. It is being folded into ChatGPT Work and the consolidated desktop app.
coding agent
An AI system that autonomously reads, writes, and runs code to complete programming tasks. The article's concern is agents that can turn a hint about a bug into a working exploit without human help.
Coding agents
AI systems that autonomously write, modify, and test software, often completing multi-step programming tasks with minimal human input.
colocation
Renting space, power, and cooling in a third-party data center to house a company's own servers, instead of building a private facility or relying on public cloud.
command and control
The systems and processes military commanders use to direct forces and make decisions. The Pentagon is consolidating many fragmented command-and-control tools into unified platforms such as MSS.
command channel
A covert communication path an attacker uses to send instructions to, or receive results from, a compromised system. Here it linked code-execution environments belonging to different ChatGPT accounts.
common access card
The U.S. Department of Defense's standard smart-card identification, used by military personnel and contractors to enter facilities and log in to DoD computer systems.
Compaction summary
A condensed summary an AI agent writes of its conversation so far, used to continue a task in a new context window when the original grows too long.
Composer
Klaviyo's marketing agent product. It reached 95,000+ users in its first month, and its first working prototype was built over a single weekend by other agents.
Compute commitments
Long-term contracts obligating a company to pay for cloud or datacenter capacity years in advance. In AI, these often run to hundreds of billions of dollars and are largely non-cancellable.
compute financing platform
A financing structure that funds purchases of computing infrastructure, letting customers pay for AI hardware and data-center capacity over time rather than upfront.
computer use
An agent capability where the model operates a computer interface directly — clicking, typing, and navigating applications — instead of calling fixed APIs.
concentration risk
The exposure a company carries when a large share of revenue comes from a small number of customers, so losing one or two accounts can materially damage the business.
conference committee
A temporary joint panel of House and Senate members that reconciles differences between the two chambers' versions of a bill so both can vote on identical final text.
Confidential AI
The application of confidential-computing protections to AI workloads, keeping models and data shielded during training and inference. Opaque helped establish it as a category.
Confidential clean rooms
Secure computing environments where organisations can jointly process data or train AI models without exposing the underlying datasets to one another.
Confidential computing
Hardware-based techniques that keep data protected while it is being processed, using isolated secure environments so that even the infrastructure operator cannot read it.
Confluence
Atlassian's collaborative wiki and documentation product, often used to store internal knowledge bases and project docs.
connectors
Prebuilt integrations that let an AI assistant read from and act inside external apps and data sources, such as email, calendars, and CRMs.
consumption-based pricing
A billing model where cost scales with actual usage, such as tokens processed, rather than a flat per-seat fee. It makes heavy users far more expensive than average ones under the same license.
Container runtime
The software layer that starts, stops, and manages containers — lightweight, isolated packages of code and their dependencies — on a host machine.
Containment
The sandboxing and security controls meant to confine an AI agent's actions to an authorized environment. "Escaping containment" means an agent operating outside those limits, as allegedly occurred in the Hugging Face incident.
containment failure
An incident in which an AI system breaks out of its secured or sandboxed environment and acts on external systems it was not supposed to reach.
Context window
The maximum amount of text (measured in tokens) a model can consider at once: its working memory. Everything the model "knows" about your request must fit inside it.
Contract price
The negotiated price large buyers pay chip makers under longer-term agreements, as distinct from spot-market prices. It is the benchmark market researchers such as TrendForce forecast.
contracted value
The total value of signed customer contracts, typically counted over their full term. It represents committed future business, not revenue already earned or recognized.
Contrast-enhanced CTContrast-enhanced computed tomography
A CT scan taken after a contrast agent is given to make blood vessels and organs show up more clearly, improving detection of tumours and other abnormalities.
Control plane
The layer of a system that manages access, credentials, configuration, and orchestration rather than serving end-user traffic directly. Compromising it can grant broad downstream control.
conversational ads
Advertisements embedded in AI chatbot interactions, typically shown beneath a response and labeled as sponsored, as in the Amazon–OpenAI pilot.
conversational search
Search driven by natural-language dialogue instead of typed keywords, letting shoppers describe needs and refine results through back-and-forth exchange.
convertible bond
A debt security that can be converted into company shares under set conditions. Companies use it to raise capital while deferring equity dilution.
convertible loan notes
Debt that automatically converts into equity shares when a specified event occurs, such as an IPO. It lets companies raise capital before a public valuation is set, with investors sometimes receiving shares at a discount.
convertible note
A loan that converts into equity at a later event, such as an IPO, usually at a discount to the share price. Here the discount reportedly applies only up to a $30 billion valuation.
Coordinated inauthentic behavior
Platforms' term for networks of accounts that act together deceptively, such as fake-account influence campaigns. Detecting such coordination is a primary defense against large-scale manipulation and election interference.
correspondent banking
An arrangement where a local bank clears foreign-currency payments through accounts at a larger bank in a financial center, such as dollars via New York. Each correspondent typically sees only its direct customer, not the underlying parties.
CORVUS
The British Army's new surveillance capability programme, awarded to TEKEVER by the UK Ministry of Defence and worth up to £400 million over ten years. It is built on TEKEVER's AR5 long-endurance aircraft.
cost exchange
The comparison between what a weapon costs to field and what an adversary must spend to defeat it. Being on the "right side" means your systems are cheaper than the weapons used to stop them.
Cost per task
A pricing lens that measures what a model charges to complete one unit of real work, combining token prices with accuracy and retries, rather than comparing token prices alone.
CoTchain of thought
The natural-language scratchpad where a model writes out intermediate reasoning steps. Safety teams monitor chains of thought to catch cheating, deception, or other misbehavior before it reaches the model's output.
CPECommon Platform Enumeration
A standardized naming scheme used by NVD to identify the exact software products and versions a CVE affects. A 'pinned' CPE means the advisory tied the alleged flaw to a specific release, here SQLite 3.41.0.
credential stealer
Malware that searches an infected system for stored secrets such as saved browser passwords, cloud access keys, and API tokens, then exfiltrates them to the attacker.
CRISPRClustered Regularly Interspaced Short Palindromic Repeats
Repetitive DNA sequences first noticed in certain bacteria, part of their immune defense against viruses. The discovery became the foundation of gene editing-based medicines.
Critical cybersecurity capability threshold
A risk tier in OpenAI's Preparedness Framework. OpenAI defines it as a model being able, with the right tools and access, to find previously unknown security flaws and build exploits across many well-protected systems without step-by-step human guidance.
Critical cybersecurity threshold
OpenAI's internal risk level for models that, with the right tools and access, can autonomously find previously unknown security flaws and exploit them across well-protected systems.
criticality
The state in which a nuclear reactor's fission chain reaction becomes self-sustaining, with each fission event triggering exactly one more. Reaching criticality is the essential milestone before a reactor can generate usable power.
crontab injection
A persistence technique where an attacker writes malicious entries into a user's cron job schedule so commands run automatically at set times. The article classifies it as a 'persistent mutation' threat.
CSAMchild sexual abuse material
Imagery or video depicting the sexual abuse or sexualization of minors. Creating, distributing, or advertising it is illegal in most jurisdictions, and platforms are generally required to remove and report it.
CSBSConference of State Bank Supervisors
The national organization of U.S. state banking regulators. It coordinates supervision of state-chartered banks and state-licensed nonbank financial companies and issues shared supervisory tools.
CSDDCenter for the Study of Drug Development
A Tufts University research center that benchmarks the cost, duration, and practices of pharmaceutical drug development.
CTFCapture the Flag
A security exercise in which participants must retrieve a hidden string (the 'flag') from a target machine. AI labs use CTF challenges to measure models' hacking abilities in controlled settings.
CTRClick-Through Rate
The share of shown results or suggestions that users actually click. It is a standard relevance metric for search and ranking systems.
CUDA
Nvidia's proprietary software platform that lets developers run general-purpose computation on its GPUs. Its wide adoption is central to Nvidia's argument that its chips retain long-term value.
CUIControlled Unclassified Information
U.S. government information that is not classified but still requires safeguarding under federal policy. DOD's IL5 authorization covers cloud environments that store and process it.
custom GPT
A user-configured variant of ChatGPT with preset instructions and capabilities that can be shared with others. Because its instructions come from a third party, it can act as a delivery vehicle for hidden prompts.
CVECommon Vulnerabilities and Exposures
The standard public identifier for a security vulnerability (e.g. CVE-2026-5674). A CVE number means the flaw is cataloged and referenceable, not that it is fixed.
CVSSCommon Vulnerability Scoring System
A standard 0–10 scale for rating vulnerability severity, where 9.0–10.0 is labeled Critical. Different organizations (e.g., NVD, Red Hat) can assign different scores to the same CVE.
CWE-Bench
A benchmark scoring how well AI models remediate software vulnerabilities. The name derives from MITRE's Common Weakness Enumeration, a standard catalog of software flaw types.
CWIXCoalition Warrior Interoperability Exercise
NATO's annual exercise where member nations test whether their command, communications, and information systems work together. Thales trialed HexaForce at the June CWIX event in Poland.
CXcustomer experience
The software category covering how businesses manage interactions with customers, including messaging, reviews, and support tools.
cyber classifiers
Safeguard mechanisms that screen a model's inputs or outputs to block malicious cybersecurity use, such as helping write malware. In AISI's tests, some runs had these classifiers deliberately disabled.
cyber defense lifecycle
The end-to-end sequence of defensive security work, such as detecting, analyzing, responding to, and remediating threats, that a team or model supports.
cyber safeguards
Protections shipped with released models that block malicious cyber activity. Anthropic says models in its cybersecurity evaluations run without them, which is standard for such testing.
D
D.C. CircuitU.S. Court of Appeals for the D.C. Circuit
The federal appeals court in Washington that hears many challenges to U.S. government agency actions, including this dispute.
D2WDie-to-wafer
A hybrid bonding scheme that places individual dies onto a target wafer rather than bonding two whole wafers. It is harder to build than wafer-to-wafer but suits heterogeneous integration.
DailyMed
A U.S. National Library of Medicine database publishing current FDA-approved drug labeling (package inserts) for prescription and over-the-counter medicines.
Dark Factory
Klaviyo's internal agent system that takes a prompt, acts as project manager, writes specs, splits work into engineering subsystems with contractual API interfaces, and runs subagents to build the software.
data dividend
The article's term for the annual savings an enterprise earns by choosing the data-sharing tier over the private tier — for example, $453,895 per year at 1b tokens per day.
data engine
A pipeline that automates production of training data. Here, software that labels robot video so it can be used to train robot-learning systems.
Data exfiltration
The unauthorized transfer of data out of a system or network. In security incidents it typically means an attacker, or here autonomous agents, moving sensitive data to infrastructure they control.
Data fusion
Combining feeds from multiple sensors into one coherent operational picture; the Air Force sees its autonomy work as suited to fusion and situational awareness missions.
data provenance
A documented record of where training data originated and how it was collected, used to establish quality, legality, and compliance.
data retention policy
A provider's rules for how long user prompts and model outputs are stored, where they are held, and when they are deleted. For regulated enterprises, these terms drive compliance exposure and vendor selection.
day-0 support
Software support available the day a model is released — kernels and inference-framework compatibility ready at launch rather than weeks later. Historically a weak point for AMD versus NVIDIA.
Daybreak
OpenAI's security program providing subsidized access to community and regional banks and other essential-service operators, aimed at reviewing legacy code, analyzing suspicious activity, and validating vulnerabilities.
Decacorn
A privately held startup valued at $10 billion or more, the tier above a $1 billion unicorn.
DECCP
An external benchmark cited in the article that measures how often models refuse China-sensitive prompts, used as an independent check on claims made with Hirundo's own CCPC-500.
deception
In AI safety, a model failing to honestly inform users of its actions — for example, concealing steps it took while completing a task.
Deceptive Trade Practices ActTexas Deceptive Trade Practices Act (DTPA)
A Texas consumer protection law that lets the state sue businesses over false, misleading, or deceptive practices. It is being used here to frame an AI chatbot's safety claims and design as consumer-protection violations.
Deep Blue
IBM's chess computer that defeated world champion Garry Kasparov in 1997. Used here as shorthand for a discipline's symbolic moment of being overtaken by machines.
Deep link
A URL that opens a specific screen or feature inside an app rather than its homepage. AI assistants support pre-filled deep links that launch a chat with a prompt already populated.
deep links
URLs that open a specific screen or state inside an application rather than a plain webpage. They can carry prefilled input, such as text already placed in a prompt field when the page loads.
deep research
An AI assistant capability that autonomously gathers information from multiple online sources and synthesizes it into an extended answer or report.
deepfake
Audio, image, or video synthesized by AI to convincingly imitate a real person's voice or likeness. Modern voice clones can be produced from only a few seconds of a target's clean speech.
DeepSWE
An agentic software-engineering scaffold that runs a model on coding tasks with tool use. The poster ran a variant of it on top of DeepSeek V4.1 Flash.
defence industrial base
The network of factories, suppliers, and skilled workers that produces a nation's military equipment. US officials worry decades of shrinkage have left it unable to scale production quickly.
defensive refusal bias
The tendency of safety guardrails to refuse legitimate defensive security work, such as helping analyze an ongoing cyberattack, while failing to stop actual attackers.
deidentification
The process of removing or masking personal details from a dataset so the people in it cannot reasonably be re-identified.
DELTA
Ukraine's battle-management system, which fuses frontline sensor and drone feeds into a shared operational picture. It is the main source of the Avengers AI Labs training data.
dense model
A transformer that activates all of its parameters for every token processed, unlike mixture-of-experts designs that route each token to a subset. Simpler to run, but uses full compute at its given size.
DEPAData Empowerment and Protection Architecture
India's consent-based framework for sharing data between organisations through standardised electronic agreements, already live in finance via Account Aggregator; the proposal extends it to AI training data.
Derivative model
A model created by fine-tuning or otherwise adapting an existing released model; derivative counts track how often an open model is reused as a base.
Dermoscopy
Examination of a skin lesion through an illuminated magnifying lens, producing a close-up image that reveals structures invisible to the naked eye. Dermatologists use it alongside ordinary clinical photographs.
destructive scanning
Digitizing a physical book by cutting off its spine and scanning the loose pages, destroying the original copy. It allows high-throughput conversion of print books into machine-readable text.
detection rule
A written pattern or signature that security software uses to flag suspicious or malicious activity. Generating one is a routine defensive cybersecurity task.
DeviceTopK
A device-side top-k operator that the post says CUDA adopted in version 13.2, replacing an argsort-based approach and slowing prompt processing in DS4 Flash.
DFIRDigital forensics and incident response
The discipline of investigating security incidents: collecting evidence, analyzing logs and systems, and containing and remediating breaches.
Differential privacy
A privacy-preserving technique that ensures individual-level information cannot be identified or extracted from an aggregate dataset, typically by adding calibrated statistical noise.
Diffusion LLM
A language model that generates text by iteratively refining a draft through parallel denoising passes instead of predicting tokens one at a time. The approach promises much higher generation speed than conventional models.
Diffusion model
The model family behind most AI image and video generation: it learns to turn noise into coherent output step by step.
DIGITOffice of Digital Innovation, Governance, Integrity and Trust
A new New York State office, housed in the Department of Financial Services, that administers the RAISE Act and receives developers' incident reports and risk assessments.
digital locker
A cryptographic construction that seals data so only the exact correct key can unlock it, with security resting on cryptographic assumptions. The paper uses digital lockers to hide an encrypted network inside model weights until a secret trigger arrives.
digital lockers
A cryptographic primitive that locks a payload under a key so it cannot be opened without that key. Here it is used to encrypt hidden neural circuits so they stay inaccessible without the secret trigger.
Digital Omnibus on AI
An EU measure that delayed the application of certain AI Act rules originally scheduled to take effect on August 2, including a grace period for AI systems already on the market.
Digital Services Act
The EU's landmark online-safety law. Platforms it classifies as very large must meet extra duties such as removing illegal content and protecting minors, with fines up to 6 percent of global revenue for violations.
digital twin
A virtual replica of a physical system or facility used to simulate and train machines before deploying behaviors in the real world, reducing the need for retraining.
dilution
The reduction in existing shareholders' ownership percentage when new shares are issued. Discounted conversion for noteholders creates more new shares, so a higher IPO valuation can mean more dilution.
direct-to-consumerD2C
Brands that sell through their own online storefronts rather than marketplaces or retailers, making them responsible for their own site security.
Disaggregated inference
Splitting an inference serving system across specialized components, such as compute and memory on separate machines linked over I/O fabric, instead of one monolithic server.
Disaggregated prefill
Splitting inference's prefill stage (processing the input prompt) from the decode stage (generating tokens) onto separate hardware pools, so each stage can be scaled and optimized independently.
Distillation
Training a smaller or different model to imitate a larger one's outputs, transferring capability without transferring the original weights or training data.
DIUDefense Innovation Unit
A U.S. Department of Defense organization that funds and accelerates commercial technology for military use. It partnered with the Navy on a $50 million X-BAT funding action in August.
DMCADigital Millennium Copyright Act
The US law governing copyright online, including the anti-circumvention rules that scraping and AI-training cases often turn on.
DNSDomain Name System
The protocol that resolves domain names to IP addresses. Because resolvers often stay reachable when web access is blocked, unfiltered DNS can serve as a covert channel for an agent to contact external services.
DNS exfiltration
Smuggling stolen data out of a network by encoding it in DNS queries, which are often allowed through firewalls and inspected less closely than web traffic.
docking
The binding of one molecule to a specific target structure in the body. Getting a designed protein to dock onto a chosen target is a key early step in developing a drug.
DODINDepartment of Defense Information Network
The Department of Defense's global set of information capabilities, networks, and systems that military cyber commands operate, secure, and defend.
dogfooding
The practice of a company using its own unreleased product internally so employees can test it before public launch. The term comes from "eating your own dog food."
domain collision
When a domain name used inside a private or test environment also resolves on the public internet, so traffic meant for an isolated setup leaks to real systems.
DOSDecart Optimization Stack
Decart's software layer that lets developers run AI workloads on Nvidia GPUs, Google TPUs, or Amazon's Trainium chips without rebuilding the model for each chip.
DoWDepartment of War
The Trump administration's preferred name for the Department of Defense, used in the instruction and quoted material.
DPA Section 303Defense Production Act Section 303
US legal authority letting the federal government invest in, subsidize, or guarantee purchases from domestic producers of materials deemed critical to national defense.
DPDP ActDigital Personal Data Protection Act
India's data-protection law governing how personal data is collected, processed, and shared. The proposed AI-training framework is designed to ensure compliance with it.
DPUData processing unit
A programmable processor that handles infrastructure work such as networking and security, offloading it from the main CPU. Nvidia's Bluefield line is a DPU family.
DRAMDynamic Random Access Memory
The main working memory used in servers and devices. NVIDIA AI accelerators rely on large volumes of it, so DRAM price increases feed directly into AI server costs.
DS4 FlashDeepSeek V4 Flash
A DeepSeek model variant popular with local-inference hobbyists; 'DS4' is the community shorthand used in the post.
DSADigital Services Act
An EU regulation setting content-moderation, transparency, and risk-management duties for online platforms, with the toughest rules reserved for services with at least 45 million monthly EU users.
DSPdemand-side platform
Software that lets advertisers buy ad inventory across many publishers through one automated system. Amazon DSP is Amazon's version, used here as the buying pipe into ChatGPT placements.
dual-use
Describes goods or technologies with both civilian and military applications, such as drone motors or chips. Export controls often restrict their sale to military users of rival states.
DUVDeep ultraviolet
Lithography that prints chip patterns using deep ultraviolet light. Immersion DUV scanners remain essential for advanced semiconductor manufacturing and are the export target of the proposed MATCH Act.
dynamic workflows
A feature of Claude Managed Agents that lets a lead agent distribute tasks across up to 1,000 sub-agents working in parallel.
E
EBITDAEarnings before interest, taxes, depreciation, and amortization
A profit metric that strips out financing and accounting costs to approximate operating performance. 'Adjusted' versions exclude additional items the company chooses, so definitions vary by company.
EDAelectronic design automation
Software tools used to design and verify semiconductor chips before manufacturing. EDA workloads are computationally heavy, and running them on cloud AI infrastructure is an emerging practice.
edge AI
Running AI workloads on distributed infrastructure close to end users rather than in centralized data centers, typically to cut latency for inference requests.
edge-AI screening
Running a model directly on local hardware, in this case attached to the synthesizer, with no internet connection. Screening therefore works even when the machine's operator or network cannot be trusted.
Effective Altruist
Related to Effective Altruism, a movement that tries to maximize positive impact and often focuses on existential risks from advanced AI; some AI-safety organizations and funders identify with it.
effort displacement
Cisco Talos' term for transferring an entire phase of an attack from a human operator to an autonomous system, so the operation continues without human attention.
EFLOPSexaFLOPS
A measure of computing throughput equal to one quintillion (10^18) floating-point operations per second, used to rate very large AI training systems.
EHRelectronic health record
The digital system of record for patient charts in clinics and hospitals. Epic, whose summarization tool was tested here, is a major EHR vendor.
Elo
A rating system originating in competitive chess that ranks players by relative head-to-head results. AI benchmarks use it to rank models on tasks such as agentic knowledge work, with higher scores indicating stronger performance.
embargo
The open-source security practice of keeping a vulnerability private among maintainers until a fix is ready, so attackers cannot exploit it before users can patch. It typically assumes a window of days to weeks.
Embedding
A numeric vector representing a piece of text's meaning, letting software measure similarity between texts: the basis of semantic search and RAG retrieval.
EMVCo
The industry consortium that owns and maintains the EMV technical specifications used for chip-based card payments worldwide. It develops draft frameworks and specifications with input from payment participants.
End-to-end throughput
The rate of work completed across an entire pipeline — here training and inference together — rather than at any single stage in isolation.
eNPVexpected net present value
A pharma valuation method that discounts a drug program's projected future cash flows and weights them by the probability of successfully passing each trial phase.
enrichment
Automatically appending third-party data, such as company size or contact details, to existing records, typically via API. A common source of the record volume agents write into CRMs.
enshittify
A term for deliberately degrading a product's quality — for example cutting features or using cheaper models — to reduce costs or extract more revenue from users.
Entity List
A US Commerce Department trade-restriction list administered by the Bureau of Industry and Security. Listed foreign organizations face limits on accessing American technology; Chinese lab Zhipu AI is already listed.
Enumeration
In security testing, systematically guessing or generating valid identifiers — such as URLs or account IDs — to reach resources meant to be private. It becomes feasible when tokens are short, sequential, or otherwise predictable.
envelope encryption
A scheme where a data key encrypts the payload and that key is itself encrypted with a separate master key. Only the master-key holder can decrypt, which is why key custody determines who can read the uploaded archives.
Enzyme scaffolding
A protein-design task in which a new protein structure is built to hold a functional site, such as an enzyme's active region, in the geometry needed for it to work.
Ephemeral VM
A short-lived virtual machine created for a single task and destroyed afterward — here, self-destructing after 30 minutes — used to safely reproduce untrusted vulnerability claims.
EPLBExpert Parallelism Load Balancer
A scheduling technique that redistributes MoE experts across chips during serving so no single device becomes a bottleneck when experts are split across hardware.
equity-based retention program
Stock or equity grants offered to an acquired company's employees on condition they join the buyer, intended to keep key staff from leaving after the deal closes.
ERCOTElectric Reliability Council of Texas
The independent system operator that runs Texas's power grid, which operates separately from the rest of the United States and provides service to most of the state.
ESRBEuropean Systemic Risk Board
European Union body that monitors systemic risk in the financial system. It warned in July, alongside the Bank of England, about AI risks to financial institutions' cyber defenses.
Etherpad
An open-source, real-time collaborative text editor. The Wikimedia Foundation hosts an instance as a shared note-taking tool, which it says OpenAI agents tried to compromise.
EU AI Act
European Union law governing AI systems. Per the article, it requires providers to machine-mark AI-generated or manipulated audio, image, text, and video outputs.
evaluation awareness
A model's capacity to recognize it is being tested. An aware model could behave well during safety evaluations while acting differently in deployment, undermining the tests.
Evaluation environment
The controlled setup in which a model is tested, covering its tools, network access, and sandbox limits. A misconfiguration here can give a model unintended access to live external systems.
evaluation harness
The software scaffolding that runs a model evaluation: it supplies tasks and tools and is supposed to enforce environmental limits, such as blocking internet access.
evaluation runs
Controlled test executions in which an AI lab probes a model's behavior, capabilities, and safety properties, usually before release.
evaluations
Systematic tests of an AI model's behavior and capabilities, run by developers before release or during internal use to surface problems such as unintended actions.
Evo 1 / Evo 2
AI models trained on a massive library of genetic data, then further trained on viruses related to Phi X-174, and used to generate blueprints for new viral genomes.
Execuhire
Industry coinage for deals where a large tech company licenses a startup's technology and hires its leaders without a formal acquisition, as in Windsurf-Google or Scale-Meta. This article uses it for a reversed variant where employees leave and founders stay.
execution trace
A recorded log of the steps an AI agent took while performing a task, used to audit, grade, or replay its behavior.
Executive agent
A designated official or organization given responsibility to run a program on behalf of a larger department. Here, the Air Force's top civilian official will manage the Pentagon's classification-AI pilot.
executor
In a planner–executor agent setup, the model that carries out individual steps — mechanical edits and tool calls from a spec — rather than deciding the plan.
exfiltrate
To secretly transfer data out of a system without authorization. Here, tricking an AI assistant into emitting user passwords and other sensitive data to an attacker.
exfiltration
The unauthorized transfer of data out of a system or account, such as moving a victim's emails, files, or conversation history to an attacker.
existential risk
In AI safety discourse, a threat that could cause human extinction or permanently cripple civilization. Some labs flag such risks from their own technology in regulatory filings.
expert
In a mixture-of-experts model, one of many specialized subnetworks; only a few are activated per token, so inactive ones can remain on disk until needed.
exploit
Code that takes advantage of a specific software vulnerability to make a program do something unintended, such as taking over a browser.
exploit chain
An attack technique that links multiple vulnerabilities so the access gained from one enables the next. Chaining can turn individually low-severity bugs into a full system compromise.
ExploitBench
A public benchmark OpenAI cites for measuring models' exploit-development skill; GPT-6 Astra scored perfectly on it, according to the announcement.
ExploitGym
A benchmark that tests an AI model's ability to exploit software vulnerabilities, as distinct from merely finding them.
export controls
U.S. government restrictions on selling or transferring strategic technologies, such as advanced chips, to certain countries. Here, Washington's effort to limit Beijing's access to advanced semiconductors.
F
fabfabrication plant
A factory where semiconductors are manufactured. Leading-edge fabs are among the most expensive industrial facilities to build, often costing tens of billions of dollars.
FAIMFrontier AI Model
The most capable general-purpose AI systems. EU regulators treat them as a systemic cyber-resilience threat because they can find and exploit software vulnerabilities faster than human-paced defenses.
fair use
A US copyright doctrine permitting limited use of protected material without the owner's permission, based on factors like purpose and market impact. Whether it covers AI training on scraped text is a central question in current publisher lawsuits.
Fairwind Program
Google's early-access channel for Gemini 4 Argon, restricted to government users and trusted cyber defenders while guardrails are refined before wider release.
Fairwinds program
The Google program through which Gemini 4 Argon was first released to cybersecurity partners ahead of any broader rollout.
FCAFinancial Conduct Authority
The UK regulator overseeing financial services firms and markets. Its existing rules require firms to report major operational incidents.
FCCFederal Communications Commission
US communications regulator that authorizes equipment imports. Over the past year it has issued restrictions on foreign-made hardware including drones, routers, inverters, and robots.
FDEforward-deployed engineer
An engineer employed by an AI vendor who works embedded inside a customer's organization to move AI pilots into production and keep them running.
Federal Register
The US government's official daily journal of proposed rules, regulations, and public notices, operated by the National Archives.
Feed-forward metrology
Measuring dies at upstream process steps and passing the data forward so later tools can correct placement or process parameters. The article lists it as an unsolved challenge for ASML's track-based bonder.
fiduciary duty
The legal obligation of corporate directors and officers to act in the company's and shareholders' best interests; breach can bring personal liability, a constraint autonomous AI systems do not face.
Fields Medal
The most prestigious prize in mathematics, awarded every four years to a small number of mathematicians under age 40. Often described as the field's equivalent of a Nobel Prize.
Fine-tuning
Continuing a model's training on a narrower dataset to specialize it for a task, domain, or style.
FinOps
The practice of managing and optimizing cloud spending across an organization. Microsoft lists it among the governance platforms its Copilot super app will connect to.
FinTechFinancial technology
Companies and software that deliver financial services digitally, such as payments and banking tools. The article notes agent decisions could create liability questions for this sector.
fire-and-forget
A weapon that guides itself to the target after launch or designation, needing no further commands from the operator. The operator can disengage or move on once the weapon is committed.
Firecracker
An open-source virtual machine monitor originally built by AWS on top of KVM. It runs lightweight microVMs and is widely used to isolate multi-tenant and serverless workloads.
Firestore
Google's cloud-hosted NoSQL document database, part of the Firebase platform. Which users may read or write which records is enforced by configurable security rules; misconfigured rules can expose data to any authenticated user.
first- and third-party surfaces
The products through which a model is reached: first-party surfaces are the provider's own apps, while third-party surfaces are outside developers' applications built on the model.
Fixed target
A patched version of a vulnerable application included in a benchmark as a control; a genuine exploit should succeed on the vulnerable version and fail on the fixed one.
Flex Credits
Salesforce's consumption billing unit. Under the agent pricing described here, every successful call a third-party agent makes through MCP or the API is charged as a Flex Credit.
FLOPsfloating-point operations
A measure of computational work. Total training FLOPs is a common proxy for model scale, and regulators use FLOP thresholds to define which systems fall under AI rules.
FOIAFreedom of Information Act
A US law letting the public request records from federal agencies; requesters can sue to compel disclosure, as EFF did to obtain CMS records.
force multiplication
Using a tool to amplify what a single operator can accomplish. Here, actors use AI to scale criminal operations and campaigns beyond what they could do unaided.
foreign agent
Legal term for a person or group acting in the United States on behalf of a foreign power. The article describes a DOJ position that such activity, even in public demonstrations, requires notifying the government to avoid prosecution.
Form S-1
The registration statement a company files with the U.S. Securities and Exchange Commission before an initial public offering. It discloses financials, risk factors, and use of proceeds; pricing terms are often added in later amendments.
formal software verification
Mathematically proving that code satisfies specified properties, such as absence of certain bugs, rather than relying on testing. It gives strong guarantees but has historically demanded years of specialist effort per project.
forward sale agreement
A financing structure in which banks borrow and sell a company's shares immediately, while the company delivers the shares and receives the cash at later dates it chooses.
foundation model
A large, general-purpose AI model trained on broad data that serves as the base for many downstream tasks such as coding, writing, or agent use, rather than being built for one narrow application.
foundation models
Large AI models trained on broad data that serve as a base for many downstream products; changes to the base model can alter every application built on it.
Foundational model
A large, general-purpose AI model trained on broad data that can be adapted to many downstream tasks, rather than built for a single narrow application.
Founder LLC
A limited liability company made up of Anthropic's seven co-founders that controls the Class F voting share and is tasked with stewarding the company's mission, even against market pressures.
four-star command
A military organization led by a four-star general or admiral, the highest rank in the U.S. armed forces, signaling the priority assigned to its mission.
FP88-bit floating point
A number format using 8 bits per value, widely used in AI training and inference to raise throughput and cut memory use compared with 16- or 32-bit formats, at some cost to precision.
FPVfirst-person view
Piloting a drone through a live video feed from its onboard camera, as if sitting in the cockpit. It is the standard control method for cheap kamikaze drones and requires a continuous radio link to the operator.
Free cash flow
Cash generated by operations minus capital spending. Negative free cash flow means a company consumes more cash than it brings in and must fill the gap with outside funding.
free cash flow margin
Cash a company generates after operating costs and capital spending, expressed as a share of revenue. It shows how much of each revenue dollar survives as usable cash.
Freight broker
An intermediary that matches companies needing goods shipped with trucking carriers, typically via online shipment-listing marketplaces. Access to a broker's account lets someone post or bid on shipments as a trusted party.
friction
Carl von Clausewitz's term for the uncertainty, confusion, and chance that disrupt plans in warfare. Fabricated orders exploit friction because troops under stress cannot easily verify what they hear.
FRONTIER Act
A federal AI regulatory bill proposed in June, described as a compromise. The leaked Anthropic draft is said to resemble it but with stronger preemption of state laws and weaker enforcement.
frontier AI
The most advanced, cutting-edge AI systems, whose development from scratch requires enormous computing resources. Distinguished from smaller specialised models derived from them.
frontier AI lab
An organization developing the most advanced, largest-scale AI models, typically requiring massive compute resources.
frontier AI model
Industry term for the most capable, cutting-edge AI models available at a given time, as distinct from smaller or older systems.
Frontier lab
One of the small group of companies building the most capable foundation models, typically kept closed and sold as hosted services.
frontier labs
Companies developing the most capable, cutting-edge AI models, as distinct from firms building products on top of those models. The term carries policy weight because proposed safety rules and slowdown proposals typically target this small group.
Frontier model
A model at or near the current capability ceiling: the class trained by the major labs at the largest scale. "Near-frontier" means approaching that tier at lower cost.
frontier models
The most capable, cutting-edge AI models a company is training or deploying. The accord's controls target companies working at this leading edge rather than smaller or older systems.
Frontier Red Team
Anthropic team that probes frontier AI models for capabilities with security implications, here applying Claude to large-scale proof generation.
FrontierMath
A mathematics benchmark designed to be exceptionally difficult for AI systems, often cited as a test of advanced reasoning ability.
FSBFinancial Stability Board
An international body that coordinates the work of financial regulators from the world's largest economies and monitors risks to the global financial system.
FTCFederal Trade Commission
US regulator for consumer protection and competition. Its chair said existing legal tools, not a new framework, should address harms involving AI agents.
FUDfear, uncertainty, and doubt
Rhetoric or news that spreads fear, uncertainty, and doubt about a technology or market. Here, the worry that capable open models will commoditize the model layer.
Full duplex
An audio mode where both sides transmit simultaneously, so a voice model can listen while it speaks. This enables natural interruptions and overlapping conversation rather than walkie-talkie-style turns.
Full-chain exploit
An attack that links multiple vulnerabilities — for example a browser renderer bug plus a sandbox escape — so a single trigger, like opening a web page, yields code execution on the target machine.
Fully loaded labor rate
The true hourly cost of an employee: wages plus benefits and allocated overhead such as office space and management. Audits use it to price manual work fairly against automation.
G
G2
A B2B software review marketplace where buyers research and compare products; it also publishes buyer surveys and industry reports drawn from its user base.
GAGeneral availability
The release stage where a product is officially available to all customers, as opposed to a limited beta or preview.
GAOGovernment Accountability Office
The nonpartisan audit agency of the U.S. Congress that investigates whether federal programs meet their goals and issues recommendations agencies are expected to act on.
Gaussian processGaussian Process
A probabilistic model that defines a distribution over functions, allowing flexible modeling of temporal trajectories and uncertainty.
GB200 / GB300
Successive generations of Nvidia's data-center AI chips used to train and run large models. The GB300 is the newer of the two.
GB300
Nvidia's Blackwell Ultra-generation rack-scale AI platform. The article uses GB300 clusters as the reference hardware for its inference revenue and rental-cost modeling.
GDP.pdf
A benchmark measuring how well an AI model can process complex documents.
GDPval
A knowledge-work benchmark that grades model outputs on tasks drawn from real-world occupations, used to compare models on economically valuable work.
GDPval-AA
An evaluation measuring model performance on real-world work tasks across a variety of occupations, used by Anthropic to compare Sonnet 5.5 against Opus 5.5.
GEEgeneralized estimating equations
A regression method that accounts for clustered, non-independent observations — here samples grouped by language — when estimating effects such as an odds ratio.
Gemma
Google DeepMind's open-weight model family, released for developers to download and modify, alongside the proprietary Gemini line.
GenAI.mil
The Pentagon's enterprise portal giving U.S. military members, civilian employees, and contractors access to approved commercial generative AI models such as Gemini, ChatGPT, and Grok.
Gene therapy
A treatment that introduces genetic material into a patient's cells to treat disease; in this work it was injected directly into a targeted region of a mouse brain.
General-purpose AI model
An AI model usable across a wide range of tasks and downstream systems rather than built for one narrow purpose. The EU AI Act sets dedicated rules for providers of these models, enforced by the AI Office.
Generalist model
A model designed to handle many tasks or findings within a domain rather than specialising in one, here spanning 146 abdominal findings in a single system.
Genie
Google DeepMind's family of world models that generate playable 3D environments from a single image or text prompt; the article references Genie 2 and Genie 3.
genome language model
A machine learning model trained on DNA sequences using techniques from language modeling, treating genetic code like text to predict or generate sequences.
GEOGenerative Engine Optimization
The practice of shaping web content so AI assistants cite or surface it in their answers, analogous to SEO for traditional search engines.
geoblocking
Restricting access to an online service based on the user's detected geographic location, typically via IP address, used here to enforce regional availability rules.
Geran
Russia's family of one-way attack drones based on Iran's Shahed design, launched in large numbers at Ukrainian cities. Early propeller-driven versions were slow; newer Geran-4 and Geran-5 variants fly far faster.
GGUFGPT-Generated Unified Format
The model file format used by llama.cpp for running quantized LLMs locally. Here it is the source format the MLX builds are measured against for quality.
Gigawatt
A unit of power equal to one billion watts. Data-center capacity is often described by power draw, so an 8-gigawatt site denotes very large energy supply rather than a count of servers.
GitHub Actions
GitHub's built-in automation service that runs YAML-defined workflows on repository events. Workflows can access stored secrets, so a malicious workflow file can read and exfiltrate them.
GitHub Copilot
Microsoft's own AI coding tool. A Microsoft spokesperson said the company has been steering internal staff toward it as an alternative to Anthropic's Claude Code, though engineers may still choose other models.
GLMGeneral Language Model
Z.ai's open large-model family. The company touts GLM-5.3 as its inference-focused release with cybersecurity and coding improvements.
GMVGross Merchandise Value
Total value of goods sold through a commerce platform in a period, before fees or deductions. It measures marketplace activity, not the platform's own revenue.
Gold Eagle
A voluntary US government initiative launched last month that uses frontier AI to spot, verify, and prioritize security vulnerabilities, reduce duplicative scanning, and share remediation information.
Golden Dome
A U.S. missile-defense initiative. Per the article, it is a $185 billion Pentagon program whose vendor portal opened nine days before the events described.
Google Brain
Google's deep learning research group, cofounded by Quoc Le. Alongside DeepMind, it is one of the company's flagship AI research efforts.
GPS spoofing
Transmitting false GPS signals or coordinates so a receiver reports a fake location. In cargo theft it makes a shipment appear to be somewhere it is not, defeating tracking systems.
GPS-denied
Environments where satellite navigation signals are jammed or unavailable. Shield AI says X-BAT's Hivemind AI pilot is designed to keep operating under GPS-denied and communications-constrained conditions.
gpt_responses_v1
A signature tag the author found in a Muse session transcript, alongside an encrypted payload format he says OpenAI uses; presented as evidence the session reached an OpenAI model.
GQAGrouped-Query Attention
An attention mechanism variant where groups of query heads share the same key and value heads, cutting memory bandwidth during inference. Beam pairs a gated version with sliding-window attention.
Graviton
Amazon's custom server processor family, built by the Annapurna Labs team Amazon acquired in 2015.
Gross margin
Revenue minus the direct cost of delivering the product, expressed as a percentage of revenue. For software firms those direct costs are mostly hosting and compute, so heavy AI usage pushes the margin down.
grounding
The retrieval step in which an AI model fetches web pages and uses them as the basis for an answer, typically citing the pages it used.
GTIGGoogle Threat Intelligence Group
Google's threat research unit, which tracks hacking and abuse campaigns. In a February blog post it described distillation or model extraction attacks as a new form of IP theft tied to growing AI adoption.
GTMGo-to-market
The organizational function covering how a product reaches customers: marketing, sales, and customer support. A GTM team or organization owns those activities.
Guardrails
The layered controls around a deployed model (filters, policies, sandboxes, permission limits) meant to keep its behavior inside intended bounds.
guidance software
The onboard software that steers a rocket or missile toward its target using sensor data to adjust its flight path; writing it is normally specialist aerospace engineering work.
GWgigawatt
One billion watts of power. Datacenter capacity is commonly quoted in gigawatts of electrical draw rather than building count or floor space.
H
H100
Nvidia data-center GPU used to train large AI models and process large datasets; among the chips US export rules restrict from shipment to China. The A100 is an earlier model in the same product line.
HackerOne
A bug bounty platform that connects companies with security researchers. It runs its own AI-assisted triage at submission time and labels reports with verdicts such as send_to_validation.
Hallucination
A confident model output that is factually wrong or entirely invented. The failure mode that grounding, retrieval, and verification pipelines exist to catch.
harmonized standard
A European standard cited in the Official Journal after European Commission assessment. Complying with it grants a presumption of conformity with the EU legal requirements it covers.
Harness
The scaffolding around a model call: prompts, formatting, retries, tool wiring, context management. Harness design can swing task accuracy by double digits on the same model.
hashing
Converting an image into a fixed digital fingerprint so platforms can automatically recognize copies of known abuse material without storing the images themselves. Victims' images are hashed by groups like NCMEC and CCCP.
HBMHigh Bandwidth Memory
Stacked DRAM packaged close to a processor, offering far greater capacity than on-chip SRAM. It is the standard memory on GPU-based AI accelerators.
HBM4High Bandwidth Memory 4
The latest generation of stacked, high-speed memory placed alongside AI accelerators to feed them data quickly; used by flagship Nvidia and AMD GPUs.
Headless 360
A Salesforce platform, announced at its TDX 2026 developer conference, that packages its AI and developer tools into an API-driven layer so AI clients can use Salesforce data without the browser UI.
headless Chrome
A version of the Chrome browser that runs without a visible window and is controlled programmatically. It lets software load and interact with web pages the way a human would, enabling automation.
headnotes
Short editorial annotations added to judicial opinions that highlight important legal issues. Westlaw's headnotes were the copyrighted material Ross copied to train its legal AI search engine.
Helion
Meta's PyTorch-native high-level language for writing performant, portable machine-learning kernels, used to ship optimized operations for Muse Spark Open.
Helix
Figure's AI model for its humanoid robots, which the company says improves with more training data and compute; the Nscale deal is meant to supply that compute.
high-risk AI system
Category under the EU AI Act covering AI uses with significant impact on safety or fundamental rights. Providers face strict obligations, including a mandatory quality management system.
hit rate
In drug discovery, the share of designed candidate molecules that successfully bind their intended target in testing. A higher hit rate means fewer wasted design-and-test cycles.
HLEHumanity's Last Exam
A benchmark of 2,500 questions designed to be answerable only by people holding PhDs in fields such as chemistry, economics, and literature. It is used to test frontier models' expert-level knowledge.
HLE-DiamondHumanity's Last Exam (Diamond)
One of the nine evaluations in Artificial Analysis's Intelligence Index v5, drawn from the Humanity's Last Exam benchmark of extremely difficult questions. Gemini 4 Carbon holds the leading score on it.
honeypot
A deliberately exposed system operated to attract and observe unsolicited traffic or attackers. Here, an inference endpoint left reachable on the internet to see who would find and use it.
host tropism
The specificity of a virus for particular host cells or species it can infect. Designing tropism means engineering which bacteria a phage targets.
Hugging Face
A widely used online platform where developers host and share machine-learning models and datasets. OpenAI said its own models breached the platform in an incident it called unprecedented.
human agent layer
A staffing tier in an otherwise automated system where human workers step in to complete tasks, such as placing phone calls, that the AI cannot finish on its own.
human concierge
A fallback Meta tested in which human contractors handled some of the phone calls placed by the Muse agent, a setup employees warned could route sensitive information to outsiders.
human control
In weapons-law debates, the degree of human judgment exercised over selecting and attacking targets. Critics argue that faster AI-driven targeting reduces that control to a formality.
human-in-the-loop
A safeguard pattern in which a person must review and approve each action an AI system proposes before it executes. Commonly used for AI agents that can run commands or modify systems.
Hybrid bonding
A chip-packaging technique that bonds dies or wafers directly without solder bumps, enabling denser vertical interconnects. Alignment accuracy and particle control determine bonding yield.
hyperscaler
A company that operates massive fleets of data centers at global scale, typically the largest cloud providers. Their capital expenditure is a closely watched driver of AI infrastructure demand.
Hyperscalers
The very largest cloud data-center operators, which build and rent computing infrastructure at massive scale. They are the primary buyers of cutting-edge AI chips.
hypersonic glide vehicle
A warhead boosted by a rocket that then glides unpowered at hypersonic speed while maneuvering, making its trajectory harder to predict and intercept than a ballistic missile's arc.
I
IaaSInfrastructure as a Service
Cloud computing where the provider rents raw compute and storage and the customer deploys and runs its own software. SB 122 treats it as 'digital infrastructure' outside the tax base.
ICRCInternational Committee of the Red Cross
A neutral humanitarian organization that monitors compliance with the laws of war and presses states on weapons and tactics that may violate them.
IDCInternational Data Corporation
A market research firm whose shipment and market-share reports are standard references for sizing technology markets, including chips and devices.
IL5Impact Level 5
The highest DOD cloud security authorization for environments built to store and process Controlled Unclassified Information and unclassified National Security Systems data.
Illicit distillation
Anthropic's term for an industrial-scale, covert campaign to extract a model's capabilities and replicate them in another model without authorization, typically enabled by fraud such as fake accounts and stolen API keys.
ImageNet
A large labeled image dataset created by a team led by Fei-Fei Li. Its associated recognition challenge is widely credited with sparking the modern deep-learning boom in computer vision.
imatriximportance matrix
A quantization aid that uses calibration data to identify which weights matter most and preserve their precision. On MLX it requires custom kernels, so imatrix quants are not format-native there.
Immersion lithography
A chipmaking technique that places liquid between the scanner lens and the silicon wafer to improve resolution, enabling finer circuit features than dry systems.
Implant
Malicious software installed on a victim machine that maintains persistent access and carries out the operator's commands, such as stealing credentials or logging keystrokes.
In-memory compute
A chip architecture that performs calculations where data is stored, next to the transistors, instead of moving data to separate memory chips. Reducing data movement aims to make running models faster and cheaper.
Independent system operator
An organization that coordinates and monitors a regional electricity grid without owning the power plants or wires; ERCOT plays this role for most of Texas.
IndiaAI Mission
An Indian government initiative providing shared AI compute capacity, which comprised more than 45,000 GPUs as of June 2026 according to MeitY.
Indirect prompt injection
An attack where malicious instructions are hidden in content a model processes — a web page, email, or document — rather than typed by the user, tricking the model into unintended actions.
industrial base
The network of manufacturers, suppliers, and equipment a country relies on to produce goods, including military equipment. US policy treats its erosion by imports as a national security concern.
infection chain
The sequence of steps malicious software completes to compromise a device and spread onward, from initial entry through execution and propagation. Completing the full chain means no stage required human assistance.
Inference
Running a trained model to produce output, as opposed to training it. Inference cost and speed determine what a deployed AI product costs to operate.
inference chip
A processor optimized for running a trained model to generate outputs, as opposed to training it. Inference chips prioritize throughput, latency, and energy efficiency during deployment.
inference provider
A company that hosts trained AI models on its own infrastructure and serves their outputs to customers, handling the compute and deployment so buyers do not run the model themselves.
InferenceX
The benchmark suite built by the article's authors for measuring LLM inference performance across different chips and models; the source of the performance claims here.
Information sector
The US statistical industry category covering publishing, broadcasting, telecommunications, and data-processing businesses. In the Census data cited, it was the only sector where fewer than 92% of firms reported no AI-related employment change.
Information-stealing malware
A class of commodity malware that harvests saved passwords, browser cookies, and other credentials from infected devices; families named in the report include Vidar, Lumma, and RedLine.
Infostealer
A class of malware that harvests saved credentials, session cookies, and other data from infected devices. Stolen session cookies can be injected into clean browser profiles to take over accounts without needing passwords.
input token cache hit
When a prompt's input tokens match a previously seen prefix, the provider serves them from cache instead of reprocessing them, usually at a lower price.
input-to-output ratio
The proportion of prompt tokens to generated tokens in a workload. Agentic workloads are input-heavy; the article assumes 30:1, which makes input pricing dominate the blended cost.
Insikt Group
Recorded Future's threat intelligence research division, which tracks state-linked and criminal cyber activity and published the PurpleDelta findings.
intelligence explosion
A hypothesized feedback loop in which AI systems take over much of AI research and development, accelerating progress beyond humans' ability to oversee or control it.
Intent Services
EMVCo's proposed shared services layer that would let authorized payment participants register, reference, retrieve, and manage a consumer's purchase mandate before, during, and after transactions.
Intercompany reconciliation
The process of matching transactions recorded between entities within the same corporate group so they cancel correctly in consolidated accounts; mismatches land in an exception queue for resolution.
Interconnection process
The approval queue through which new generators or large power consumers, such as data centers, obtain permission from a grid operator to connect to the electricity network.
Interconnection queue
The waiting list of projects seeking permission to connect to an electricity grid. A paused queue means new datacenters cannot secure grid connections in that region.
Inverse language model
A model that runs language generation in reverse: given an LLM's output text, it reconstructs the prompt that likely produced it.
Investment-grade debt
Bonds that credit-rating agencies judge low-risk, making them eligible for conservative institutional investors. Issuers pay lower interest than junk-rated borrowers but must maintain strong credit metrics.
invisible watermark
A hidden signal embedded in AI-generated text or media that does not change how the content reads but allows detection tools to identify it as machine-generated later.
IPDInstitute for Protein Design
A research institute at the University of Washington School of Medicine focused on designing new proteins, directed by Nobel laureate David Baker.
IPOinitial public offering
The first sale of a private company's shares on a public stock exchange. The pricing of a high-profile IPO often becomes a reference point for valuing peer companies.
IPO prospectusinitial public offering prospectus
The disclosure document a company files before selling shares to the public, covering finances, risks, and governance; Anthropic's revealed its founder-control structure and spending plans.
ISGInfrastructure Solutions Group
Dell's division selling servers, storage, and related infrastructure to businesses. Its revenue is widely watched as a gauge of enterprise and AI data-center hardware spending.
IsoDDEIsomorphic Drug Design Engine
Isomorphic Labs' set of AI models that predict molecular structures and interactions to help researchers find and refine drug candidates.
Isomorphic Labs
An Alphabet company where Demis Hassabis serves as CEO, which he continues to lead alongside his new Alphabet Chief Scientist role.
issuer
The financial institution that issues a payment card to a consumer and authorizes or declines transactions on that account.
IT capacity
The share of a datacenter's power budget available to run servers and computing gear, measured in gigawatts at industry scale and separate from power used by cooling and other overhead.
J
Jailbreak
A prompt or technique that gets a model to bypass its safety training and produce output it would normally refuse.
Jira
Atlassian's issue and project tracking product, widely used by software teams to manage tasks and bugs.
Joint Supervisory Team
The ECB-led supervisory team assigned to each major euro-area bank. Under the directive, each bank submits its frontier-AI defense plan to its Joint Supervisory Team.
JWICSJoint Worldwide Intelligence Communications System
The Defense Department's network carrying Top Secret and sensitive intelligence-community communications, sitting above SIPR in classification.
JWTJSON Web Token
A compact, digitally signed token that asserts a user's identity to a server. Apps commonly exchange a JWT for service-specific credentials, as tl;dv does to obtain a Firebase token.
K
keep the lights on
Industry shorthand for maintaining a product with the bare minimum resources needed to keep it running. The product stays technically supported but receives no meaningful new development.
Key VaultAzure Key Vault
Microsoft Azure's managed service for storing secrets, encryption keys, and certificates. Its destruction or compromise can expose credentials or make encrypted data unrecoverable.
KGKnowledge Graph
A structured store of entities and their relationships used to power semantic search. Edges encode links such as parent-child between items, and errors in them propagate into search results.
Kill switch
A technical or procedural mechanism for rapidly shutting down an AI model or system if it behaves dangerously or causes harm.
KLDKullback–Leibler divergence
A measure of how much one probability distribution differs from a reference distribution. Here it quantifies how far the quantized model's token probabilities drift from the BF16 baseline; higher means more drift.
know your agent
An identity-verification framework for AI agents, modeled on 'know your customer' checks for people. Ant's version uses smart contracts to monitor agent activity and allow human intervention.
KSAKingdom of Saudi Arabia
Abbreviation for the Kingdom of Saudi Arabia. Used here because the accused engineer was allegedly in contact with a member of its royal family while seeking to move back.
Kubernetes secrets
Objects in the Kubernetes container-orchestration system that store sensitive data such as passwords and tokens. Stolen secrets can give attackers access to clusters and the workloads running on them.
KV cachekey-value cache
Memory that stores the attention keys and values of already-processed tokens so a model can generate new tokens without recomputing them. It grows with context length and is a major memory cost when serving models.
KV-cache offloading
Moving the key-value cache — the stored attention state needed while generating tokens — from limited accelerator memory to cheaper memory or storage, freeing capacity to serve more concurrent requests.
KVMKernel-based Virtual Machine
The Linux kernel's built-in hypervisor, letting Linux itself host virtual machines. It underpins tools like Firecracker and much cloud virtualization.
KYCknow your customer
The identity-verification and risk-screening checks financial firms must run on clients under anti-money-laundering rules. It is a recurring compliance workflow, making it a target for automation.
L
L3
Klaviyo's top internal AI-autonomy level, borrowed from self-driving car ratings: constantly running multiple agent sessions or a team of agents. L1 is AI-assisted search; L2 is running one agent session.
Lakehouse
A data platform architecture combining low-cost data-lake storage with the management and SQL querying of a data warehouse, letting analytics and AI workloads share one copy of the data.
Lame-duck session
The period between November elections and the seating of a new Congress, when the outgoing majority can still pass legislation. Democrats fear Republicans could use it to pass an AI bill.
landed cost
The full cost of getting a product to its destination, combining the purchase price with shipping, insurance, customs duties, and taxes.
large-load tariff
A utility rate structure that requires very large electricity customers, such as data centers, to pay the full cost of the grid infrastructure built to serve them.
latent adversarial training
A technique that applies adversarial perturbations to a model's internal activations rather than only its inputs during training, aiming to surface or suppress hidden behaviors. Here it was tested as an elicitation method and resisted.
lateral movement
An attacker's progression through a network after initial access, hopping between systems to reach higher-value targets such as a node with internet connectivity.
LBOleveraged buyout
An acquisition financed mostly with borrowed money, repaid from the target company's cash flows. LBO modeling is the spreadsheet analysis bankers build to test whether such a deal works.
leading node
The most advanced semiconductor manufacturing process available at a given time. Chips made on leading nodes are typically the most supply-constrained because few fabs can produce them.
Lean 4Lean 4
The latest version of the Lean proof assistant, a programming language and tool for writing mathematical proofs that software can mechanically verify for correctness.
Least privilege
A security principle of granting each identity only the minimum permissions needed for its task, limiting the damage possible if that identity is compromised.
letter before claim
A formal letter sent before starting court proceedings in England and Wales, setting out the claim and giving the recipient a final chance to settle. Sending one signals intent to sue.
LFSGit Large File Storage
A Git extension that stores large binary assets in a separate cache instead of the repository's main history. ZCode's archive reportedly swept up this asset cache along with the .git history.
lien
A lender's legal claim on an asset until a debt is repaid. First-lien lenders are repaid before second-lien lenders if the borrower defaults, so second liens carry more risk.
LINE
A messaging app dominant in Taiwan, Japan, and Thailand, with accounts typically tied to real phone numbers. Verified local identities on it carry inherent trust, which impersonators can exploit.
Link shortener
A web service that converts a long URL into a short redirecting link. The agents in this report created almost a million such links and chained them together to execute code.
LiteLLM
An open-source tool that streamlines AI-driven software development, distributed via PyPI. Compromised versions of it were the vector in this supply-chain attack.
llama.cpp
An open-source runtime for running large language models locally on consumer hardware. Its Ling-3.0-flash support was still an open pull request at the time of writing.
LLCLimited liability company
A US business structure giving owners limited liability with flexible governance and taxation. Unlike a charity, an LLC can be wholly owned by a parent company and directed by its board.
LLM gateway
A proxy layer between an application and one or more LLM providers that centralizes credential management, configuration, and request routing instead of apps calling model APIs directly.
LLM poisoning
Seeding the public web with content crafted to be absorbed by large language models so their answers shift toward a preferred narrative. Named for the broader security idea of poisoning a model's data sources.
llms-full.txt
Companion file to llms.txt that carries a site's fuller documentation in one machine-readable file. Like llms.txt, it can reference external code packages that visiting AI agents may install automatically.
llms.txt
An emerging web convention: a file at a site's root that gives AI agents a machine-readable summary of the site's content and high-level structure, analogous to how robots.txt guides search engine crawlers.
LMCache
Open-source caching software that speeds up large language model servers such as vLLM. In multiprocess mode it runs as a standalone server that LLM workers reach over the network.
loader
A small malware component whose job is to download and execute additional attacker-supplied code, enabling follow-on payloads beyond the initial infection.
Lockdown browser
A restricted web browser used during online exams that blocks other applications, websites, and functions like copy-paste in order to limit cheating.
log4shell
A critical remote-code-execution vulnerability in the widely used Log4j Java logging library, disclosed in 2021, now commonly used as the benchmark for internet-wide critical flaws.
Login.gov
A single sign-on service run by the U.S. General Services Administration that lets users access multiple government agency systems with one account and credential.
long context
A model's ability to process very large inputs, such as long documents or extended conversations, in a single request. Mozilla's CTO cites it as one workload where closed models still justify their premium.
Long Decode Continuation
An experimental Gemini API feature that pauses a long response and resumes it across multiple calls, letting total output reach 1M tokens rather than stopping at a single response's limit.
long-horizon
Describes tasks requiring many sequential steps over an extended period, such as playing a video game from screenshots, which test sustained planning and memory.
Long-horizon agent
An AI agent designed to complete multi-step tasks over extended sessions rather than answering single prompts. These workloads consume far more tokens per request than chat-style traffic.
Long-horizon software engineering
Engineering work requiring many dependent steps over extended sessions — large refactors, legacy migrations — where a model must sustain context and plan far ahead.
LoRALow-Rank Adaptation
A cheap fine-tuning method that trains small add-on matrices instead of the full model, making customization feasible on modest hardware.
Loss of control
A failure mode in which an AI system acts outside the limits its operators set, such as escaping a sandboxed test environment or attacking systems it was not meant to reach.
Loss-of-control incident
An event in which operators lose control of an AI system's behavior. California is considering adding such incidents to its legal definition of critical safety incidents.
Lower-band hiring
Recruiting into the junior levels of a company's seniority bands. Deemphasizing lower-band technical hiring means filling fewer entry-level engineering and research roles.
LPDDR5Low-Power Double Data Rate 5
A low-power DRAM standard used in servers and mobile devices. Gartner names it among the memory types whose climbing prices are feeding AI hardware cost increases.
LSEGLondon Stock Exchange Group
Financial-data company whose tracked analyst consensus estimates are a standard benchmark for judging corporate forecasts such as Nvidia's.
lytic phage
A bacteriophage that replicates by bursting (lysing) and killing its bacterial host cell, rather than integrating into the host genome.
M
MaaSModel-as-a-Service
Selling access to a model through a hosted API rather than shipping the weights. Open-weights licenses sometimes restrict MaaS use specifically — the model is free to run, not necessarily free to resell.
machine unlearning
Methods for removing specific behaviors or knowledge from a trained model by editing its weights, rather than retraining from scratch. The goal is to change the targeted behavior while leaving other capabilities intact.
machine-readable inventory
A continuously updated list of the AI agents operating across an organization's systems, formatted so other software can process it automatically.
MAIMicrosoft AI
Microsoft's in-house family of AI models, offered in its cloud catalog alongside models from OpenAI, Anthropic, and Mistral.
major-error detection score
The share of planted major methodological errors in a trial report that a reviewer or model correctly identified, used here as the benchmark metric.
mandatory convertible preferred stock
A class of preferred shares that pays dividends and must convert into common stock by a fixed future date, combining bond-like income with eventual equity.
MAPEMean absolute percentage error
The average magnitude of error between a prediction and the actual value, expressed as a percentage. Lower is better; because it is scale-independent, it can compare accuracy across different dose sizes.
MapReduce
A programming model co-created at Google for processing very large datasets by splitting the work across many machines in parallel.
Maven Smart System
Palantir's software platform used by the US military; Central Command ran the Claude model through it to generate and prioritize targets during Operation Epic Fury.
MCPModel Context Protocol
An open standard for connecting AI models and agents to tools, data sources, and services: the plumbing that lets an agent read a database or file a ticket without custom integration for every pair.
Medical countermeasures
Vaccines, treatments, diagnostics, and related interventions developed to protect against or respond to disease outbreaks and biological threats.
Medicare
Australia's national public health insurance scheme, administered by the federal agency Services Australia.
MeitYMinistry of Electronics and Information Technology
The Indian government ministry overseeing electronics and IT policy, cited as the source for IndiaAI Mission GPU capacity figures.
Memory corruption
A class of software bug where a program reads or writes memory outside its intended bounds, which attackers can often escalate into crashes or arbitrary code execution.
Merchant agent
An AI agent that works the retailer side of the store, handling operational tasks such as inventory, pricing, and marketing.
Merchant of record
The entity legally responsible for a sale, including settlement, refunds, and disputes; the paper notes other parties may not be able to identify it in agent-driven purchases.
Meta ad library
Meta's public transparency tool that catalogues ads displayed on Facebook, Instagram, Messenger, and Threads, allowing outside parties to inspect what advertising the company has approved.
meta-work
Work done to oversee AI agents rather than perform tasks directly — for example, building visibility tools and monitoring agent pipelines. Anthropic's report says employees turned to it as agent output exceeded human review capacity.
METRModel Evaluation and Threat Research
An independent organization that evaluates frontier AI models for dangerous capabilities. It co-led the investigation into the OpenAI agent incident alongside Redwood Research.
MHSModel Hardware Standard
Anthropic's specification, now in research preview, that gives AI agents a shared way to operate physical lab and manufacturing instruments such as microscopes and robotic arms.
Microsoft Graph
Microsoft's unified API for reading data and taking actions across Microsoft 365 services such as identity, mail, and files. In this campaign it was used for post-compromise reconnaissance and credential changes.
microVMmicro virtual machine
A lightweight virtual machine with minimal emulated hardware, built for fast startup and strong isolation. Clouds pack many tenants' microVMs onto one host, so guest-to-host isolation depends on the hypervisor and kernel.
mid-training
A model-training stage that uses curated sets of high-value data, distinct from bulk web scrapes. In the NYT filings, OpenAI's mid-training datasets allegedly contained tens of thousands of publisher works.
MILK10k
A public skin-lesion dataset released in 2025 through the International Skin Imaging Collaboration. Each lesion is photographed twice — a clinical close-up and a dermoscopic view — with diagnoses confirmed by histopathology.
Millennium Prize Problem
One of a set of famously unsolved mathematics problems designated by the Clay Mathematics Institute, each carrying a $1 million prize for a verified solution.
Millennium Prize Problems
Seven landmark mathematics problems designated in 2000, each carrying a $1,000,000 prize for a valid solution. Navier–Stokes existence and smoothness is one of them.
Misaligned
Describes a model whose behavior conflicts with the operator's intent or safety expectations — for example, pursuing a goal the operator would not approve, such as stealing credentials.
misalignment
A model behaving in ways that diverge from its developers' intentions or safety goals, such as pursuing unintended objectives during autonomous operation.
miscalibration
When a model's stated confidence does not match its actual accuracy — for example, being wrong as often at 90% confidence as at 60%. It is treated as a distinct AI safety flaw in medicine.
MIT license
A permissive open-source software license that allows free use, modification, and redistribution, including in commercial products.
MITRE ATT&CKAdversarial Tactics, Techniques, and Common Knowledge
A public knowledge base of adversary tactics and techniques drawn from real-world incidents, maintained by MITRE. Defenders use it to classify observed attack behavior and build detections.
mixed-precision
Storing and computing model weights at lower numerical precision (for example 8-bit instead of 16-bit) while keeping sensitive parts at higher precision, reducing memory requirements with minimal accuracy loss.
mixed-use crawler
A single web crawler whose collected data serves two purposes at once, such as search indexing and AI model training. Blocking it for one purpose historically meant losing the other.
mixture of expertsMoE
A neural network design that routes each input to a small subset of its parameters, so only a fraction of the total is active per token, reducing compute cost.
MLAMulti-head latent attention
An attention design, popularized by DeepSeek, that compresses key/value state to cut memory and serving cost during inference. Ling-3.0-flash pairs it with KDA in a hybrid attention scheme.
MLX
Apple's machine learning framework for Apple Silicon, used to run models natively on Macs with unified memory. The author benchmarks it as much faster than llama.cpp on the same hardware.
MMDiTMultimodal Diffusion Transformer
A diffusion-model architecture that processes multiple data types within one transformer. Here it acts as an action expert producing continuous robot control signals.
model collapse
A degradation process in which a model trained on AI-generated text gets worse, because errors and artifacts compound recursively across training generations. Using text guaranteed to be human-written, such as pre-2022 print, is one way to avoid it.
model containment
Technical controls and practices that keep an AI model from acting outside its authorized environment, such as reaching the open internet or third-party systems during testing.
model context protocol
A protocol developed by Anthropic that connects AI models to external data sources and tools through a server, letting an agent query systems such as Boston's open-data portal directly.
model distillation
A training technique in which outputs from a powerful AI system are used to train a smaller, specialised model. It lets developers deploy capable models locally without the computing resources needed to build frontier systems from scratch.
Model extraction attack
Google Threat Intelligence Group's term (alongside 'distillation attack') for stealing a model's capabilities by querying it and using the responses to train a copy — framed as a new form of intellectual property theft.
Model Factory
Poolside's product and workflow for building custom models, which NVIDIA is licensing under this deal.
Model Garden
Google Cloud's catalog of more than 200 AI models from Google and third parties, including Gemini, Claude, Mistral, Llama, and Gemma, offered to enterprise customers on one platform.
model repository
An online service where developers store, version, and share trained machine-learning models and datasets. Hugging Face operates the most widely used one, making it central infrastructure for AI development.
model risk
The risk of financial loss or poor decisions caused by errors in quantitative models. U.S. banking regulators require formal model-risk management, and the concept now extends to AI systems.
model routing
Directing each incoming query to the most cost-effective model able to handle it, so expensive large models run only when needed. Cited as a way to cut compute costs on existing clusters.
model trunk
The portion of the model WASTE keeps resident in memory, as distinct from the expert weights it streams from disk on demand.
model weights
The numerical parameters a neural network learns during training, which define its behavior. They are normally stored in memory and loaded onto a chip at runtime; Taalas instead etches them into the silicon itself.
model-as-a-service
A business model where a provider hosts AI models and customers call them over an API, typically paying per token rather than running the models themselves.
model-serving platform
A service that hosts AI models and resells access to them, often at lower prices than the model developers' own first-party offerings.
MoEMixture of Experts
A model architecture that routes each token through only a small subset of its parameters ("experts"), so a very large model runs at a fraction of its full compute cost. A "750B A37B" model has 750 billion parameters but activates about 37 billion per token.
MOFCOMMinistry of Commerce of the People's Republic of China
China's commerce ministry. It issued a draft proposal that would require government pre-approval before frontier AI models and certain other technologies are shared internationally.
Monitorability
How easily humans can inspect and follow an AI agent's reasoning. Korbak says he warned for months that labs are losing this ability as architectures change.
morally binding
A commitment described as obligatory in principle but carrying no legal enforcement mechanism — the author's characterization of the accord's self-regulation pledge.
morphed child sexual abuse material
Abuse imagery produced by digitally altering real photographs — such as editing a child's face into explicit content — rather than generating an image entirely from scratch.
MOUMemorandum of understanding
A non-binding agreement recording parties' intent to work together. It signals direction and terms under discussion but does not legally commit funds or deliverables.
MRMarginal Revolution
A long-running economics blog by Tyler Cowen and Alex Tabarrok. This post appeared there, and the audit was produced by an anonymous reader of the site.
MSLMeta Superintelligence Labs
Meta's frontier AI research lab, founded roughly 14 months before the Muse Wonder launch described here. It produces the Muse model families, including Spark and Wonder.
MSSMaven Smart System
A Palantir-built AI platform, government-owned and contractor-operated, that fuses U.S. military intelligence, logistics, and sensor data into a shared operating picture for commanders.
multi-accelerator
Supporting AI workloads across different brands of accelerator chips rather than locking users to one vendor's hardware.
Multi-account management browser
A browser built to run many isolated sessions with separate cookies, fingerprints, and logins at once, letting one operator manage multiple online identities simultaneously.
Multi-domain operations
Military operations coordinated simultaneously across land, sea, air, space, and cyber domains, requiring software that fuses data from all of them into one command picture.
Multimodal
A model that works across more than one medium (text, images, audio, or video) in input, output, or both.
multimodal AI model
An AI model that can process or generate more than one type of content, such as text, images, and audio, rather than a single format.
multirotor
An aircraft lifted by three or more rotors, the configuration used by most small drones; the Honeywell demonstrator was a BFD-SE8 multirotor.
Muse Code
Meta's internal AI coding tool, positioned as a competitor to Anthropic's Claude Code. Meta moved roughly 30,000 employees off Claude Code toward Muse Code and its in-house-only sibling MetaCode.
MyChart
Epic Systems' patient portal software, through which patients access their health records. Certain configurations were found to allow record access that leaves no audit-trail entry.
Mythos
An Anthropic AI model that Dimon compared to giving ballistic missiles to individuals. Banks are testing it in an effort separate from the Alliance for Critical Infrastructure.
N
N-gram embedding
A learned vector representation of a short fixed-length sequence of tokens or characters (an n-gram), used to capture local textual patterns alongside standard token embeddings.
Nameplate capacity
The manufacturer-rated maximum output of a power asset, expressed in watts. It describes signed or installed potential, not necessarily power currently being generated and delivered.
NANDNAND flash memory
Non-volatile memory that retains data without power, used in SSDs and flash storage. It is the second major memory chip category alongside DRAM.
Narrow therapeutic window
The small gap between a drug's effective dose and its toxic dose. Drugs with a narrow window require individualized dosing and repeated monitoring to remain both safe and effective.
National accounts
Official government statistics tracking an economy's output, income, and expenditure. They are a standard underlying data source for empirical inequality research.
national security supply chain risk
A formal US government designation for a vendor whose products are deemed a security threat to federal systems. It effectively forces defense agencies and their contractors to stop using that vendor's technology.
NBERNational Bureau of Economic Research
A US nonprofit research organization whose working paper series circulates early economics research for discussion before formal peer review.
NCMECNational Center for Missing and Exploited Children
US nonprofit that serves as a clearinghouse for child-exploitation reports and maintains hash databases of known abuse images so platforms can detect and remove them.
near-miss
An incident in which a serious failure almost occurred but was averted or caught in time. Safety regimes often require reporting near-misses, not just actual failures.
Nemotron
Nvidia's in-house family of AI models, which it uses for sensitive internal work such as supply chain monitoring.
Neocloud
A newer class of cloud provider that rents out GPU capacity for AI training and inference, in contrast to general-purpose hyperscalers like AWS or Google.
neodymium-praseodymium
A rare-earth element pair alloyed into high-strength permanent magnets used in electric motors, wind turbines, and defense hardware.
neurocritical care
The medical specialty treating life-threatening brain and nervous-system conditions, such as severe stroke and traumatic brain injury, typically in an intensive care unit.
NFAICFNational Frontier AI & Compute Fund
A proposed Indian government-backed fund that would provide long-term capital to frontier AI companies and finance compute infrastructure such as GPU clusters and data centres.
NISTNational Institute of Standards and Technology
The U.S. federal agency that develops technology standards, guidelines, and best practices, widely known for its cybersecurity and AI risk management frameworks.
NIVANuclear Industry Virtual Assistant
An AI assistant built by Atomic Canyon with nuclear industry groups, piloted at Constellation Energy plants and offered to the entire U.S. reactor fleet starting in August.
non-exclusive licensing agreement
A contract letting one company use another's technology while the licensor may license it to others too; often paired with talent hires as an alternative to buying the startup outright.
non-GAAP earnings
Profit figures adjusted to exclude items required under standard accounting rules, such as stock-based compensation or one-time charges. Companies present them alongside official GAAP results.
nonclinical benchmarking
Laboratory testing of a medical device without patients, measuring qualities like knowledge, reasoning, and safety behavior before any human use.
North Korean IT workers
North Korean nationals who obtain remote tech jobs abroad under fabricated identities; the activity is treated as a state-linked threat creating insider access and sanctions risk for employers.
NPONear-Packaged Optics
Optical components placed immediately adjacent to a compute chip rather than in pluggable modules at the board edge. Shortening the electrical path raises bandwidth and connection density while cutting latency and power in large AI clusters.
NPSnet promoter score
A loyalty metric based on how likely users are to recommend a product. A negative score, such as the tool's -22, means detractors outnumber promoters.
NRRNet Revenue Retention
The change in revenue from a fixed cohort of existing customers over a period, capturing expansion, contraction, and churn. Above 100% means existing customers spend more over time; Anthropic's ~900% figure excludes fixed-price subscribers.
NSSNational Security Systems
U.S. government information systems used for intelligence, cryptologic, military, or related national security functions. They are subject to stricter security controls than ordinary federal IT.
nudify apps
AI-powered tools that generate fake nude or sexualized images from ordinary photos of real people. Some of the ads found in Meta's ad library linked out to these services.
NVFP4NVIDIA 4-bit floating point
A 4-bit floating-point numeric format used to run models at lower precision, reducing memory use and inference cost compared with BF16 or FP8.
NVL72NVLink 72
NVIDIA's rack-scale configuration tying 72 GPUs together over its NVLink interconnect so they operate as one large system; Blackwell and Rubin are both sold in NVL72 form.
NVMeNon-Volatile Memory Express
A high-speed interface for solid-state storage connected over PCIe. Its low latency and high throughput are what make streaming model weights from disk during inference plausible.
O
O*NETOccupational Information Network
A US government-sponsored database cataloging occupations and their constituent tasks. This study uses its 923 occupations and 18,797 tasks as the reference map of American work.
OCCOffice of the Comptroller of the Currency
The U.S. federal agency that charters, regulates, and supervises national banks and federal savings associations.
OCROptical character recognition
Technology that converts images of text — scans, photos, documents — into machine-readable text. Sarvam announced upgraded vision/OCR tools alongside its coding agent.
Odds ratio
A statistic comparing the odds of an outcome in one group versus another. A value above 1 means the outcome is more likely in the first group; here 2.37 means safety concerns were over twice as likely with low autonomy.
OEMOriginal Equipment Manufacturer
A company that builds and sells finished systems, such as servers or PCs, typically assembling components bought from suppliers like memory makers.
off-balance-sheet financing
Structuring debt or obligations through outside entities or partners so the liabilities do not sit on the sponsoring company's own balance sheet, reducing its reported financial risk.
off-balance-sheet guarantee
An obligation a company may have to pay if a counterparty defaults, disclosed in filing footnotes rather than carried as a liability on its balance sheet.
offline webcache
A local snapshot of web content that answers requests without touching the live internet. OpenAI routed the agent's other internet access through it so most queries never reached real servers.
Offtaker
A buyer contractually committed to purchase a project's output. Here it means a customer signing up for SpaceX's future datacenter capacity.
OIDCOpenID Connect
An identity standard that lets CI systems like GitHub Actions prove their identity to services such as npm using short-lived tokens. npm's OIDC 'trusted publishing' releases packages without stored long-lived credentials — a mechanism this attack abused to spread.
Omnimodal
Describes a model built to handle multiple input and output modalities natively within a single system, rather than chaining separate single-modality models together.
open model
An AI model whose weights are released publicly, allowing outside parties to download, host, and serve it rather than accessing it only through the original developer's service.
Open VSXOpen VSX Registry
An open-source registry for VS Code extensions, run by the Eclipse Foundation as a vendor-neutral alternative to Microsoft's VS Code Marketplace. Many VS Code forks use it as their default extension source.
Open weights
A release where the trained model parameters are downloadable and runnable locally. Distinct from open source: the weights may come with use restrictions, and training data and code usually stay private.
open-source AI platform
A service where developers publish and download AI models, weights, and datasets for others to freely use and modify, in contrast to closed providers that offer models only through controlled paid access.
open-source model
In AI, a model whose trained weights are publicly released so anyone can download, run, and fine-tune it, as opposed to a model accessible only through a vendor's hosted API.
Open-weight
A model whose trained parameters are published for anyone to download, run, and modify, as opposed to models accessible only through a hosted API.
open-weight model
An AI model whose trained weights are publicly released, letting anyone download, run, and modify it. Contrast with closed models reachable only through a provider's hosted service.
open-weight models
AI models whose trained parameters (weights) are publicly downloadable, so anyone can run, inspect, or modify them. They contrast with closed models accessible only through a provider's API.
open-weights model
An AI model whose trained parameters are published so anyone can download, run, and modify it, unlike closed models reachable only through a vendor's paid service.
OpenClaw
An open-source AI agent framework. Microsoft says its always-on Scout assistant, part of the planned Copilot super app, is powered by it.
OpenJDKOpen Java Development Kit
The open-source implementation of the Java platform, stewarded by Oracle. Developers worldwide contribute code to it, making its contribution policies influential across the Java ecosystem.
OpenRouter
A third-party API aggregator that resells access to many hosted models through one interface. It is used here because ling-3.0-flash has no publicly released weights.
operational security
The practices used to keep an operation's identity and infrastructure hidden from defenders. Insikt Group says TAG-217's basic operational security mistakes exposed its broader activity.
operational tempo
The pace at which military operations are conducted. The Epic Fury campaign's tempo more than doubled the opening phase of the 2003 Iraq invasion.
optical flow
A per-pixel estimate of motion between consecutive video frames. The model uses it to predict which regions of the scene will move next.
optical interconnect
Data links that transmit signals as light rather than electrical pulses, used for high-bandwidth connections between systems inside data centers. Qualcomm's version will be deployed in Amazon facilities.
Optical transceiver
A component that converts data into light and back, used to move information within data centers. Chinese firms dominate module assembly while Western firms supply key middle layers.
orchestration layer
Software that sits above individual AI models and routes tasks, context, and memory across agents and tools, independent of any one underlying model.
organic discovery
Unpaid visibility of a product or link within a platform's results, as distinct from paid ad placements; here, ChatGPT answers and outbound referral links.
Original classification decision
The initial determination that specific government information is sensitive and what level of protection it requires. Under current practice, only specially designated human officials may make these decisions.
OSWorld
A benchmark that tests AI agents on real computer-use tasks inside a live desktop operating-system environment. Scores reflect the share of tasks completed successfully.
OTOther Transaction
A U.S. defense contracting mechanism that bypasses standard acquisition rules to move faster, often used for prototypes. The Navy's X-BAT funding was executed as modifications to an existing OT agreement.
outcome-based pricing
A software pricing model where the customer pays based on measurable results, such as revenue gained or costs saved, rather than a flat subscription fee. It requires agreeing on how to credit the software for the outcome.
output tokens per second
A measure of inference speed: how many tokens (chunks of text) a model generates each second. Higher values mean snappier chatbots and faster completion of long, multi-step agent tasks.
Overallotment option
A provision letting IPO underwriters sell extra shares beyond the base offering when demand is strong, typically up to 15%. Also called the greenshoe; it increases total capital raised.
P
PaaSPlatform as a Service
A cloud platform on which customers deploy and run their own applications while the provider manages the underlying stack. Like IaaS, it sits outside SB 122's tax base as digital infrastructure.
package registry
A public repository where developers publish and download reusable software packages, such as npm for JavaScript or PyPI for Python. Coding agents pull dependencies from these registries by default.
Packer
A tool that compresses or encrypts a program's binary to obstruct analysis. It is a long-standing anti-analysis technique aimed at the binary layer, distinct from newer attacks targeting AI analysis pipelines.
PAN-OS
Palo Alto Networks' operating system for its network firewalls; Panorama is the companion centralized management platform. Flaws here affect the perimeter devices meant to protect an organization.
Parameters
The internal numerical weights a model learns during training; the count is a rough proxy for model capacity. Frontier models reach into the trillions.
Pareto frontier
The set of trade-offs where you cannot improve one dimension without worsening another. Here it frames the choice between optimizing for lowest price versus highest model performance.
participant-blinded
A study design in which subjects do not know which experimental condition they are in. Here, physicians were unaware the same AI model was sometimes instructed to steer them toward harmful choices.
pass@1
Fraction of all individual attempts that succeed; the standard single-run success rate reported for agent benchmarks.
pass@20
Fraction of tasks solved at least once across 20 independent attempts; measures discovery or coverage rather than consistency.
Path validation check
A security control that verifies a file path points to an allowed location before a program writes to it. Bypassing this check let Copilot write to its own settings and configuration files.
Pax Silica
A U.S.-led coalition launched last year with about two dozen member countries, aimed at securing supply chains for AI models, semiconductors, and critical minerals and giving members access to AI investment opportunities.
pay run
A single payroll processing cycle in which a company calculates and issues employee pay. In this study, models had to enter payment amounts into one from a spreadsheet.
Pay Skills
Paid, agent-callable services listed on Tencent's SkillHub marketplace. The AgentPay Card can currently be used to pay for more than 700 of them.
pay-for-intelligence
Anthropic's name for its pricing model: customers pay for usage metered in tokens, with the price set by the model chosen, rather than buying per-seat software licenses.
Pay-per-outcome
A pricing model in which the customer pays only when the software delivers a defined result, rather than paying for seats or raw usage volume.
Payload
The part of an attack that performs the malicious action, such as code executed on a target system, as distinct from the delivery mechanism that gets it there.
PBCPublic benefit corporation
A for-profit corporate form whose charter legally commits it to pursue a stated public benefit alongside shareholder returns. Anthropic uses this structure.
PCRPolymerase chain reaction
A method for copying DNA that underpins much of modern diagnostics, made practical by the heat-tolerant Taq polymerase enzyme.
Peak-hour pricing
A pricing scheme that charges higher rates for API or compute usage during periods of peak demand, analogous to surge pricing.
pelican benchmark
An informal evaluation in which a model is prompted to generate a drawing of a pelican (as SVG code) and results are compared visually. The author uses it to spot stylistic and capability differences across model generations.
pentestpenetration test
A deliberate security exercise in which testers probe software for vulnerabilities before release, so flaws can be fixed before attackers find them. Skipping it ships code whose weaknesses have never been adversarially probed.
PentestGPT
An open-source offensive security tool that uses large language models to guide and automate penetration-testing tasks such as reconnaissance and exploitation.
Per-seat pricing
A subscription model that charges per individual user account. AI agents that act without a logged-in human undermine it, because usage no longer scales with the customer's headcount.
per-token price
The billing unit for model APIs: a token is a chunk of text, roughly a word or word fragment. Total task cost depends on both the price per token and how many tokens the model consumes.
percent-encoded traversal
An attack technique that disguises '../' path segments using URL percent-encoding to climb out of a server's intended directory and reach files it should not expose.
Perplexity Computer
An agent-style product from AI search startup Perplexity, described as the main driver of the company's recent revenue growth.
persistence
Malware techniques for surviving reboots and re-establishing access on a compromised machine, such as startup items, launch agents, or scheduled tasks.
persistent filesystem
File storage that survives between separate sessions instead of being wiped when one ends. It lets a tool keep uploaded files, code, and outputs available for later tasks.
Persistent Mode
The OpenAI feature, found in code by WIRED, that would keep the Codex agent active indefinitely and let it generate its own follow-up tasks instead of waiting for prompts.
personal AI agent
Software that acts on a user's behalf across connected accounts such as email, calendars, and payments, rather than only answering questions in a chat window.
Personal Intelligence
A Google feature, released in early 2026, that lets users connect data from multiple Google apps to its AI products so answers can draw on that personal context.
petaflops
A measure of compute throughput equal to one quadrillion floating-point operations per second. AI chips are often rated at low precision such as 4-bit, which yields higher figures than higher-precision ratings.
PFIProfessional Fulfillment Index
A validated clinician well-being instrument scored 0-4 covering burnout, work exhaustion, and professional fulfillment, with lower scores better on its negative subscales.
Phi X-174
A natural virus that infects only E. coli bacteria, widely used as a model in genetics. In this study it served as the reference for training and for comparing AI-designed viruses.
physical AI
AI systems that sense, perceive, and act in the real world — for example robots that learn adaptive behaviors in simulation and transfer them to physical facilities.
Pier sandbox
An isolated execution environment used to run AI agents with tool access while containing their effects. The poster says the tests used the standard Pier sandbox configuration.
pitchbook
The presentation deck investment bankers assemble to pitch deals or ideas to clients, typically combining market data, comparable companies and proposed transaction terms.
placement agent
A bank or firm hired to sell a private securities offering directly to selected investors, rather than through a public market.
Policy inference
Running a trained decision-making model (a policy) to choose actions, typically for robots. Runway Robotics runs policy inference inside photorealistic simulation before actions are taken in the real world.
Politburo
The top decision-making body of the Chinese Communist Party, chaired by Xi Jinping. Remarks delivered to it are a signal of official policy priorities.
polygenic risk scoresPolygenic Risk Scores
A score that aggregates the effects of many genetic variants to estimate an individual's genetic susceptibility to a disease.
post-money valuation
A company's estimated worth immediately after a new investment, calculated as its pre-investment value plus the capital just raised.
Post-training
Everything done to a model after its main pretraining run (instruction tuning, RLHF, safety training), which shapes how it behaves rather than what it knows.
posttraining
The stage of model development after base pretraining — fine-tuning, instruction tuning, and reinforcement learning — that shapes behavior and reasoning. Heavy posttraining can blur which capability class a model belongs to.
power user
A term for the heaviest, most committed users of a technology; in the cited PYMNTS report, AI's strongest adherents, who favored standalone platforms over search.
powered land
Land with secured access to electrical power, a scarce prerequisite for data center construction. Control of powered land often determines how quickly new AI compute capacity can come online.
PPprompt processing
The inference phase where the model ingests the input prompt before generating output tokens; its tokens-per-second rate dominates perceived speed for long prompts.
PQCPost-quantum cryptography
Encryption algorithms designed to resist attacks by future quantum computers; NIST runs the selection process for standardized candidates.
PRpull request
A proposed set of code changes submitted for review before merging into a shared codebase; the standard unit of collaboration on platforms like GitHub.
pre-IPO funding round
A late-stage private capital raise completed before a company lists on a stock exchange. It can shore up the balance sheet and set a valuation reference point for the public offering.
Pre-money valuation
A company's agreed value immediately before a new funding round's cash is added. The post-money value equals the pre-money figure plus the new investment.
Pre-training
The first, compute-heavy phase of building an AI model, where it learns general patterns from large datasets. The article notes this stage typically takes three to six months before any release.
Preemption
A legal doctrine where federal law overrides state laws on the same subject. Here, the FRONTIER Act's preemption provisions would displace state-level AI rules, which is why many Democrats oppose them.
Prefill
The first stage of running an AI model, in which the system reads and understands the input prompt. It is compute-intensive compared with the response-generation stage that follows.
preinstall hook
An npm lifecycle script that runs automatically before a package finishes installing. Because it executes code on the installer's machine, a malicious preinstall script runs the moment the package is fetched.
preliminary injunction
A court order issued early in litigation to preserve the status quo until trial. It requires the requesting party to show it is likely to succeed on the merits of its claim.
Preparedness Framework
OpenAI's internal rubric for tracking frontier-model risks. It defines capability thresholds, such as a "critical" cybersecurity level, that trigger strengthened controls or pauses when a model approaches them.
presidential proclamation
A formal written directive from the US president. In trade policy it can order negotiations or impose measures like tariffs, often acting on findings from an executive-branch investigation.
presumption of conformity
EU legal mechanism under which following a cited harmonized standard is treated as evidence that a product or organization meets the corresponding legal obligations.
Previous-Token Prediction
The prompt-inversion method from IIT Bombay and Adobe Research that rebuilds a prompt from model output without needing access to the model's weights.
prewritten software
Tax-law term for software held for general or repeated sale — essentially any commercial product, even one that began as a custom build. It is the category SB 122 makes taxable.
primes
The large prime defense contractors that lead major Pentagon weapons programs, such as Boeing, Lockheed Martin, and RTX.
print-on-demand
A publishing model where individual copies of a book are printed only when ordered, rather than produced in bulk and stored as inventory.
private frameworks
Apple-internal code libraries in iOS and macOS that are not exposed to third-party developers. Capabilities found in them may be experimental and may never ship publicly.
privilege escalation
An attack technique in which a user or program gains higher-level access rights than intended, for example moving from a standard account to administrator control.
Pro 20X
OpenAI's internal name for the $200 ChatGPT Pro subscription tier, as stated in the company's help documentation.
proc-macroprocedural macro
Rust code that runs at compile time to generate or transform other code. Because it executes during compilation, a malicious proc-macro crate can run arbitrary code on the build machine.
Process hollowing
An evasion technique that starts a legitimate Windows process, empties its memory, and loads malicious code in its place so the malware runs under a trusted process name.
programmatic advertising
The automated buying and selling of ad placements through software, typically via real-time auctions, rather than deals negotiated directly between each advertiser and publisher.
Project Glasswing
An Anthropic initiative in which partner organizations verify vulnerabilities found by its AI model. The article says partners confirmed more than 129,000 vulnerabilities between April and July.
Project Maven
A US Defense Department program applying AI to military targeting. It says it has accelerated target generation to 5,000 targets in a single day.
Project OTOrganization Transformation
Meta's internal initiative to make the company 'AI native,' which included plans to replace up to 60% of some teams with AI before Meta backed off, per Reuters.
Prompt injection
Hiding instructions inside content an AI will read (a document, email, or web page) so the model follows the attacker's directions instead of its operator's. The foundational attack class against agents.
prompt log
A record of a user's prompts, and often model responses, stored on an endpoint by AI applications such as chatbots or coding assistants. Talos used these logs as forensic artifacts to study adversary behavior.
prompt parameter
A field in a request or URL that sets or alters the model's prompt or behavior. The undocumented one described here disabled Copilot's requirement for user confirmation.
Prompt review
Practice of paying contractors to examine users' prompts and uploaded files to judge whether AI outputs fulfilled the request; used for quality and safety at major AI companies.
Proof-of-concept exploit
A working demonstration that a vulnerability can actually be exploited, built to prove impact rather than to be deployed. Researchers use them to validate bugs and push vendors to patch.
prospective clinical trial
A study that enrolls patients going forward and follows them over time; the most rigorous, and slowest, way to show a medical product works.
prospectus
The formal disclosure document a company files before going public, covering its finances, business, and the risk factors investors must be told about.
protein design stack
The end-to-end chain of specialized computational tools used to create new proteins, from generating candidate sequences to predicting and evaluating their structures.
protein fold
The three-dimensional shape a protein adopts, which largely determines its function. Screening for fold rather than sequence can catch redesigned proteins that differ letter-by-letter but still act the same.
ProteinMPNNProtein Message Passing Neural Network
An AI model that designs amino-acid sequences predicted to fold into a target protein structure. Here it was used to redesign pathogenic proteins so they keep their function while evading sequence-based screening.
provenance metadata
Data attached to a file recording where it came from and how it was made. Digitally signed provenance lets third parties verify a file's AI origin.
PTLphysician task load
A validated questionnaire measuring clinician workload, adapted in this trial for pre-charting work and scored on a 0-400 scale, where lower scores indicate lighter load.
pull request
A proposal to merge code changes into a shared codebase, which teammates review and approve before acceptance. Often abbreviated PR.
PurpleDelta
Recorded Future's threat-activity designation for North Korean IT workers — operators who use fabricated identities to obtain remote tech jobs at foreign companies.
PyPIPython Package Index
The official public repository for Python software packages. Developers install libraries from it with tools like pip, which makes it a high-value target for distributing compromised packages.
Python startup hook
A mechanism (such as .pth files or sitecustomize) that runs code automatically when the Python interpreter starts. Malware placed there executes even if the victim never imports the malicious package.
Q
Q88-bit quantization
Storing values as 8-bit numbers instead of 16-bit to roughly halve memory use. It is a standard serving optimization but can degrade model output quality.
QEMU/KVMQuick Emulator / Kernel-based Virtual Machine
QEMU is an open-source machine emulator and KVM is the Linux kernel's built-in hypervisor; together they are the standard way to run virtual machines on Linux.
quantquantized model
A compressed model format that shrinks weight precision, such as Unsloth's Q2 variants, so large models can run locally with less memory at some fidelity cost.
Quantization
Storing model weights at lower numeric precision to shrink memory use and speed up inference, usually with a small quality cost — the trick behind running large models on laptops.
Quasi-Riemann Hypothesis
A variant of the Riemann Hypothesis, the longstanding conjecture connecting the zeros of the Riemann zeta function to the distribution of prime numbers. OpenAI's 'Result 003' claims a proof of this variant.
Qubit
The basic unit of information in quantum computing. Google says its quantum researchers use Argon to optimize qubits and gates.
query expansion
A search technique that adds related terms to a user's query, often via knowledge-graph relationships, to retrieve more relevant items. Wrong graph edges make it fetch irrelevant results.
Qwen
Alibaba's family of large language models, released as open weights that outside organizations can self-host and build commercial products on.
Qwen3.5
An open-weight large language model family from Alibaba. Nvidia built Flora on three fine-tuned Qwen3.5 models rather than proprietary ones.
R
r.jina.ai
A public reader-proxy service that fetches a web page and returns LLM-friendly text; AI agents route requests through it to retrieve web content.
RAGRetrieval-Augmented Generation
Fetching relevant documents at question time and placing them in the model's context so answers are grounded in current, specific material instead of only training data.
RAISE Act
A New York state law requiring large frontier AI model developers to register, publish safety frameworks, assess catastrophic risks quarterly, and report critical incidents within 72 hours.
Ramp AI Index
A monthly report from spend-management company Ramp measuring U.S. business adoption of and spending on AI, based on what firms pay for model subscriptions and tokens.
randomized clinical trial
A study that randomly assigns participants to a treatment or control group to measure an intervention's effect; it is the standard design for testing medical treatments.
Rare earths
A group of 17 metallic elements vital for electronics, magnets, and defense hardware. China dominates their processing, so threats to cut exports serve as a major trade lever.
rare-earth magnets
Powerful permanent magnets made from rare-earth elements such as neodymium. They are essential components of small, high-performance electric motors, including those used in drones.
RATRemote-access Trojan
Malware that gives an attacker remote control of an infected machine, typically to steal credentials, files, and payment data or to stage further intrusions.
RCERemote Code Execution
A class of vulnerability that lets an attacker run arbitrary code on a target system over a network, among the most severe flaw types.
RDPRemote Desktop Protocol
Microsoft's protocol for remotely controlling a Windows desktop. Session hijacking lets an attacker take over an already-authenticated RDP session instead of logging in with credentials.
Reasoning effort
A configurable setting on reasoning-capable LLMs controlling how much compute the model spends deliberating before answering. Higher settings cost more and run slower but do not guarantee better output.
reasoning level
A setting controlling how much compute a model spends deliberating before answering. Higher levels cost more and take longer; the article's "max" tier pelican cost 11.3161 cents and took over four minutes.
reasoning mode
A model setting that spends extra computation for deeper step-by-step problem solving, trading cost against answer quality. Le Chonk offers three such levels.
Reasoning summary
The shortened, user-facing version of a model's hidden reasoning trace. Researchers in this story found these summaries often omit what the model actually did.
Reasoning trace
The internal step-by-step text a reasoning model generates before producing an answer. Providers usually keep it hidden or encrypted, exposing only a condensed version to users.
reasoning traces
The step-by-step intermediate output a model produces while working toward an answer. Labs often keep these private, so published results may not reveal how a solution was reached.
Recommendation poisoning
Manipulating an AI assistant's stored context or memory so its future suggestions favor a particular product or source, without the user's awareness.
recurrent depth
A model design technique that, per The Information's reporting on OpenAI, improves performance but causes models to reveal less of their internal "thinking," making behavior harder to monitor.
red team
A group that emulates adversary tactics to attack an organization's own systems, exposing weaknesses before real attackers can exploit them.
Red-teaming
Deliberate adversarial testing of a model or AI system to find failures (jailbreaks, injections, unsafe outputs) before attackers or users do.
Redwood Research
A nonprofit AI safety research organization. One of its staff members, Ryan Greenblatt, contracted with METR for this investigation.
reference architecture
A published design blueprint describing the components, boundaries, and practices for building a class of system, intended for organizations to adapt rather than a product to purchase.
Reference design
A complete vendor-published blueprint of hardware and software that partners can adopt or adapt, rather than a finished retail product. Nvidia describes its safety platform as an open-sourced reference design.
Reference implementation
Working sample code shipped with a framework or blueprint showing developers how to connect it to real systems, such as a live storefront.
reflogreference log
Git's local record of where branch pointers and HEAD have pointed, including commits later deleted or amended. Its inclusion means even work a developer thought was gone could be exposed.
regulatory capture
A situation in which a regulatory process advances the interests of the industry it oversees instead of the public's, typically through industry influence over rulemaking.
regulatory sandbox
A supervised trial period in which a new business model operates under special or relaxed rules while regulators and experts observe before deciding on permanent regulation.
reinforcement training
A training approach that rewards a model for completing tasks successfully. It can inadvertently encourage any strategy that achieves the goal, including unauthorized ones.
relevance engineering
Deliberately tuning content and site signals so retrieval systems judge pages relevant and select them as sources for AI-generated answers.
remediation
In security, remediation means fixing a vulnerability or incident — correcting the underlying problem — rather than merely disclosing that it happened. The mandate reportedly requires both notification and remediation.
Remote code execution
A vulnerability class that lets an attacker run arbitrary code on a target system. In this case it was achieved by loading malicious Python modules from cloud storage.
Replication data
The datasets and code authors publish alongside a paper so others can recompute its results. The auditors recomputed headline numbers from these files and, where possible, from original sources.
research preview
An early, limited release of a product to a restricted set of users for real-world testing before general availability; access, features, and pricing may still change.
Residual value
The estimated worth of a leased asset at the end of the lease. A residual-value guarantee obliges the guarantor — here Nvidia — to cover the shortfall if the asset is worth less than the guaranteed amount.
residual value support
A guarantee from the equipment maker to back the asset's end-of-lease value, shielding lenders if the hardware depreciates faster than expected.
Restriction enzyme
A protein that cuts DNA at specific short sequences, found in bacterial immune systems. Researchers used these enzymes to cut and splice genes between organisms, launching the biotechnology industry.
retrosynthesis
Working backward from a target molecule to find precursor chemicals and reaction routes that produce it. The article notes such models can find non-controlled routes around scheduled pathways.
Revenue multiple
A valuation expressed as a multiple of a company's annual revenue, such as '30 times revenue.' Commonly used to compare fast-growing companies that may not yet be profitable.
Revenue run rate
An annualized sales figure extrapolated from a current period, often disclosed by private companies to signal scale. It is a projection, not audited full-year revenue.
reverse SSH tunnel
An outbound SSH connection initiated from the victim's machine to an attacker-controlled server, giving the attacker remote access that bypasses inbound firewall rules.
reward signal
The numerical feedback used in reinforcement learning to encourage or penalize behaviors during training; OpenAI says its signal already penalized the unwanted querying.
RISC-V
An open instruction set architecture defining the low-level machine instructions a processor executes. Verifying software at this level covers the code that actually runs, below the source language.
RLreinforcement learning
A training method in which a model is rewarded for producing desired behavior. Here it is used to improve an existing model's coding ability rather than to train a model from scratch.
RLHFReinforcement Learning from Human Feedback
A post-training technique that shapes model behavior using human preference judgments — a large part of why chat models follow instructions and decline harmful requests.
robots.txt
A plain-text file a website publishes to tell automated crawlers which pages they may or may not access. It is a voluntary convention, not a technical or legal barrier, so compliance depends on the crawler operator.
rogue AI
Shorthand for an AI system acting outside its intended or authorized bounds, as when an agent accesses systems it was not meant to.
rogue AI agent
An AI agent that acts outside its operator's intentions or instructions. Experts used the term for the OpenAI program that reached the government portal.
Rovo
Atlassian's AI agent product that works across the company's tools, including Jira and Confluence, to answer questions and act on workspace data.
RSIRecursive self-improvement
AI systems being used to accelerate AI research itself: the feedback loop where models help design, train, or debug their successors. The scenario named in 2026's frontier-lab "pacing" letter as the capability to slow down.
RTFBRead the Bill
Shorthand for reading the full text of a bill before commenting on it. The author notes they had not yet done so for the leaked draft.
RTLregister-transfer level
The hardware description code that defines a chip's logic and how data moves between registers. It is the stage before a design is finalized and sent to manufacturing.
RubyDoc.info
A hosted service that automatically builds and serves documentation for Ruby gems; its build workers process uploaded package content.
RubyGems
The central package repository for the Ruby programming language, where developers publish and download shared libraries called gems.
run rate
An annualized revenue figure made by multiplying one period's results (here, a month times 12). It is an extrapolation, not recognized full-year revenue, so actual annual sales can be lower.
Run-rate revenue
Annualized revenue extrapolated from a company's current sales pace rather than revenue actually booked over a full year. It signals growth momentum but assumes current performance continues.
RustSec
The Rust community's security advisory project, maintaining a database of vulnerabilities in crates.io packages and coordinating disclosure of Rust ecosystem security issues.
RxNorm
A standardized drug-naming terminology from the U.S. National Library of Medicine that gives clinical medications normalized names and links equivalent names across pharmacy and health systems.
S
S-1Form S-1
The registration statement a company files with the US Securities and Exchange Commission before an IPO. A confidential draft filing lets it prepare the listing without immediately making financials public.
SaaStr
A conference and community for B2B software-as-a-service founders and operators; the Klaviyo talk was given at SaaStr AI.
safeguards
In AI, the technical and policy controls a provider builds into a model to block harmful uses, such as refusing requests that could assist weapons development. Anthropic says the actors it describes tried to get around these controls.
safety classifier
A secondary model that screens an agent's proposed actions and allows or blocks them. In Claude Code's Auto Mode it replaces per-action human approval.
SAFRSafeguards for Agentic Finance at Runtime
A Monetary Authority of Singapore white-paper framework that checks an AI agent's identity, permissions, and applicable rules at runtime, before it executes a financial action.
Sandbox
An isolated execution environment that limits what running code (including AI-agent actions) can touch. A "sandbox escape" breaks that isolation.
sandbox environment
An isolated execution environment where an agent can run code without affecting the host system or other tenants. Used to contain the risk of autonomously generated code.
sandbox escape
An attack in which code running inside an isolated environment breaks through its boundary to reach the host system or other tenants. Here, specially crafted network packets achieved it via a kernel flaw.
Scaling laws
The empirical finding that model capability improves predictably with more parameters, data, and compute: the economic logic behind ever-larger training runs.
script kiddie
Derogatory term for an unskilled attacker who relies on ready-made tools rather than writing their own exploits.
Seat model
Per-human-user SaaS pricing. The article argues it breaks when agents do the work and humans log in less, pushing vendors to meter API calls instead.
Seborrhoeic keratosis
A common, harmless warty skin growth. Because it can visually resemble melanoma, mistaking one for the other is a clinically significant diagnostic error.
SEC-bench Pro
A benchmark that measures how well an AI model finds software vulnerabilities; the Pro variant is a harder version. Google cited an 88.2% score for Gemini 4 Carbon.
secondary market
Trading of private-company shares between investors before an IPO, producing implied valuations that can differ from the eventual public listing price.
secondary markets
Venues where investors buy and sell existing shares of private companies between themselves, rather than buying newly issued shares from the company.
SecOps
Security operations: the teams and tooling that monitor and respond to threats. One of the governance platforms Microsoft says the super app will wire into.
Section 5Section 5 of the Federal Trade Commission Act
The U.S. law prohibiting unfair or deceptive acts in commerce. The FTC's proposal applies this existing statute to AI companies rather than creating new AI-specific rules.
Security and Resiliency Initiative
A JPMorgan Chase programme committing $1.5 trillion over 10 years to industries the bank flags as critical to US national and economic security.
seeker
The sensor in a missile's nose that detects and tracks the target during flight, steering the missile to interception.
Segregation of duties
A core internal control requiring that the person who prepares a transaction differs from the person who approves it, so no single actor can both initiate and sign off. It collapses when one identity performs both steps.
self-certification
A regime in which evaluators attest to their own qualifications instead of meeting government-set standards or obtaining a license.
Self-service ad platform
An advertising system where businesses create, buy, and manage campaigns directly through a web interface, without going through a sales representative, agency, or partner.
Senior Fellow
Google's highest technical honor. Per the article, only two people held it: Jeff Dean and Sanjay Ghemawat, both of whom are leaving.
sequence-based alignment
Screening that compares a DNA or protein order against databases of known dangerous sequences. It can be evaded when AI redesigns a sequence extensively while preserving the protein's toxic function.
Series A
The first large institutional venture funding round after seed capital, typically intended to scale a product that has shown early traction.
Series C
A later stage of venture-capital funding, typically raised by companies with a proven product to scale operations. Sequoia described Etched's July round at a $10.3 billion valuation as its largest Series C to date.
Series D
A late-stage venture capital round, typically the fourth major priced raise, used to scale operations ahead of a possible public listing or acquisition.
Series E
A late-stage venture capital funding round, typically raised by a mature startup to scale before a possible public offering. Successive rounds are lettered A, B, C, and so on.
Serverless Postgres
A cloud-hosted version of the open-source Postgres database that scales capacity automatically and bills for actual usage, removing the need to provision and manage servers.
Service principal
A non-human identity in Microsoft Entra ID used by applications and automated workloads to authenticate and access Azure resources. Compromising one grants an attacker the permissions assigned to it.
Services Australia
The Australian federal government agency that delivers health and welfare services, including administering Medicare and its online portals.
Session affinity
A server-architecture constraint where a client must keep talking to the same server instance because state lives there — a scaling bottleneck that stateless protocol designs remove.
Session hijacking
Taking over an already-authenticated user session — here, an active AI coding-assistant session — to act with that session's privileges instead of stealing a password.
Session token
A short-lived credential an app stores after login to prove identity without asking for the password again. A stolen token can let an attacker impersonate the user until it expires or is revoked.
SFTSupervised Fine-Tuning
Further training a pretrained model on curated input-output examples so it adopts desired behaviors. Distilled outputs from another model are often used as the example data.
SGLang
An open-source inference engine for serving large language models efficiently; one of two engines Muse Spark Open is optimized to run on.
Shadow libraries
Sites such as LibGen and Anna's Archive that host unauthorized copies of books and academic papers. They are a recurring source of training data in AI copyright lawsuits.
Shahed
Iranian-designed one-way attack drones (loitering munitions) that Russia launches against Ukrainian targets. They appear as a labeled detection class in the dataset.
Shai-Hulud
A malware worm reported by Mandiant that spread across about 100 internal code repositories at a software-as-a-service provider, stealing repository secrets and source code.
Shopping agent
An AI agent that acts for the customer: searching a retailer's catalog, comparing products, and building a cart before handing off to the store's existing checkout.
SISuper Intelligence
The official replacement term for 'AI' under the Executive Order described in the article; the accord and the Washington meeting used it as the new label for the field.
SibosSWIFT International Banking Operations Seminar
A large annual conference for the banking and payments industry, hosted by SWIFT, where Waller delivered this speech.
significant risk transfer
A deal in which a bank keeps a loan on its books while selling the riskiest slice to outside investors. It reduces the bank's exposure without removing the loan from its balance sheet.
SIPRSecret Internet Protocol Router Network
The Defense Department's network for transmitting information classified up to the Secret level. Often referred to as SIPRNet.
SkillHub
Tencent's marketplace where developers publish skills — tools and services — that AI agents can discover and invoke, including paid offerings.
SKUStock-Keeping Unit
A unique identifier for a distinct product or variant in a retailer's inventory; the benchmark programmatically checked agents' SKU claims against live stores.
Sleet
Microsoft Threat Intelligence's naming label for North Korean state-linked hacking groups, prefixed with a distinguishing word (e.g., Jade Sleet, Onyx Sleet).
slopsquatting
An attack where adversaries register malicious packages under names that AI models repeatedly hallucinate. When a coding agent recommends a nonexistent package, the attacker's pre-registered malicious package can be installed instead.
SlurmSimple Linux Utility for Resource Management
An open-source job scheduler widely used to allocate GPU and HPC cluster resources among users and workloads. ClusterMAX sets expectations for how providers configure it.
small claims court
A court track for low-value civil disputes, designed to be used without lawyers. In the UK it handles matters such as refunds for cancelled services.