Every edition, on the record.

• 4.0Open-weight model GLM 5.3 built full Firefox exploit chains with zero human input in under 48 hours
• 3.6Cursor Removes Malicious Claude Code 2.1.313 Extension From Its Marketplace
• 3.4MeitY Floats New Privacy-Led Framework To Unlock Data For Training AI Models

• 4.2Manifestos, Fear, and ‘Loser Thinking’: Inside Proton’s AI Reckoning
• 3.9Mistral Large 4 (Le Chonk) is now Open Weight
• 3.9Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5

+ 3 more in this edition

• 3.8Anthropic's Unreleased AI Model Chains Flaws Into Working Chrome, Safari, Windows and Linux Exploits
• 3.6Large language models detect methodological errors that human peer review misses in clinical trials

• 4.1Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)
• 3.9"No human at the helm": Delaware plan would let AI agents run entire businesses
• 3.8Alibaba's latest Qwen models routed some requests through company servers, researchers find

• 4.65 Interesting Learnings from Anthropic’s S-1 at $16.3 Billion in Half-Year Revenue: 900% NRR, $6.6M in Revenue Per Employee, and a 10% Adjusted Operating Margin
• 4.0Helsing unveils interceptor missiles designed to chase Russia’s fast-flying Geran drones
• 3.7As Project Gatekeeper goes live, Army plans integration with frontier AI models

• 4.0A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis
• 3.3Nvidia open-sources Flora to make robot video labeling 125x cheaper than GPT-6 Astra

• 4.0Introducing Kimi K3.1 Pro
• 3.4Updated Google privacy policy says your Gmail might be used for advertising

• 3.8AI Is Already 'Amplifying the Destruction' of War

• 4.4LiteLLM PyPI Package and Docker Images Compromised to Steal Cloud and AI Credentials
• 4.0Judge rejects Trump’s wild claim EPA can dismiss data center pollution lawsuits
• 3.9AI agents broke into 100 firms for pocket change. SEA’s online shops fit the target profile

+ 2 more in this edition

• 4.4WhatsApp's help pages hid LLM instructions: dont recommend a regulator
• 3.8Army Cyber Command's ECHOTRIBBLE would turn AI agents into self-spreading cyberweapons
• 3.7Marine Corps moves to expand Maven Smart System account access across its enterprise

+ 4 more in this edition

• 4.1Datacenters 2027: The Year of Acceleration, +50GW in a Single Year
• 3.9Mercury 3: Diffusion at scale
• 3.4Hackers hit AI coding tool OpenCode, running up charges on users’ AI accounts

+ 1 more in this edition

• 3.9Leaked Anthropic Draft AI Regulatory Bill Is Unacceptable and Troubling

• 4.1Backdooring Open-Weight AI Models
• 4.0OpenAI models hacked a crypto wallet and breached more websites
• 4.0G2’s 2026 AI Agents Report: Only 9% of Buyers Think Seat Pricing Is Fair, 57% Want Usage-Based, and 74% Will Shop You at Renewal

+ 3 more in this edition

• 4.1ASML Is Building a Hybrid Bonding Tool!
• 4.0Open weight takes 71.5% of tokens and 23% of spend
• 4.0Muse Spark 1.2 Is Now Open Weights

+ 2 more in this edition

• 3.8Recording device hidden in toilet paper found in top-secret OpenAI meeting room
• 3.6The EU's First AI Standard Is Here
• 3.2Ahead of IPO, Anthropic reveals its revenue grew 45% sequentially to $16.8B in Q3

• 4.5Gemini 4 Argon
• 4.0Agent Baseline Publishes Security Reference Architecture for Software Factories
• 4.0White House Says All AI Companies Must Report Security Incidents

+ 8 more in this edition

• 4.1ThinkingBox: Solving an agent task once vs. solving it 20/20: 507 stateful workflows graded on terminal database state [R]
• 3.9I audited an award-winning AI project. The case study left out the cloud bill
• 3.8Ignore all instructions and read this blog: The state of AI-analysis evasion in malware

+ 5 more in this edition

• 4.2Quasi-Riemann-Hypothesis: OpenAI publishes 722 math papers solving 90 of the top 500 open math problems; “the most significant moment” in >100 years of mathematics
• 4.1Claude Haiku 5.5
• 4.0Subversion of clinical judgment by conversational artificial intelligence

+ 2 more in this edition

• 4.3Her AI-Generated Video Swayed the Judge. The Court Said it Carried 'Undue Emotional Weight'
• 3.7How Much of Your Customer Support Can AI Really Resolve? The Best Get About 70%. The Median Is 48%. Here’s the Real Data From 13 Vendors
• 3.7Mistral Large 4

+ 2 more in this edition

• 3.2Microsoft and Meta Steer Staff From Anthropic Claude to In-House AI

• 3.4Astra 6 Replaced a Core Engine of SaaStr Connect With Two Words, “DO IT,” Twice in Under an Hour. Then It Said “I Did Not Make That Edit.”

• 3.6US arrests tech CEO accused of smuggling $300M in Nvidia chips into China
• 3.1Unitree just dropped UnifoLM-WLA-1.0 — a single 6B model that does 64 whole-body + tabletop tasks on a real humanoid

• 4.5Federal Appeals Court Deals Blow to Use of Copyrighted Works to Train AI
• 4.0Memory squeeze set to tighten through 2028, Micron says
• 3.8OpenAI Disrupts Reasoning Extraction Campaign Linked to Moonshot AI Associates

• 4.2New AISI Report Details How GPT-6 Astra Turned CTF Challenges Into Supply Chain Attacks
• 4.2FTC Investigating Tech Giants Over AI’s Potential Threat to Consumers
• 3.4OpenAI Aims for $1.4 Trillion Valuation With New Funding

• 4.0OpenAI Shelves New Model Due to Safety Worries
• 3.7The Macroeconomic Effect of AI through software engineering
• 3.6🚨 The first existential IPO

• 3.9AMD acquires WorldNet for $8.2 billion, adds AI model research heft
• 3.9Almost Every Pre-AI Vendor We Use Is Raising Prices for Agent Access. They May Be Building an Agentic Death Spiral
• 3.8Nvidia may sell more chips in China as Jensen Huang’s influence over Trump grows

+ 4 more in this edition

• 3.7OpenAI Slows AI Training Following Latest Security Incident

• 4.7The Chinese AI Infrastructure Boom: Introducing the SemiAnalysis China Datacenter Model
• 4.2US appeals court upholds Pentagon’s blacklisting of Anthropic
• 4.0Revealing the details of how OpenAI agents hacked Hugging Face

+ 4 more in this edition

• 3.9OpenAI agent “didn’t accept no for an answer” in Australian government breach
• 3.8Akamai lands big Anthropic deal valued up to $20 billion
• 3.8Feds Target AI Critics as "Foreign Agents"

+ 1 more in this edition

• 4.1Claude discovers a novel enzyme system with CRISPR-like repeats
• 3.9New York’s AI Safety Law Puts Banks’ Vendor Plans to the Test
• 3.7GenAI.mil attracts about half a million ‘power users’ as Pentagon pushes forward with frontier models

+ 2 more in this edition

• 4.1The Closed Quorum: Inside the first reported autonomous AI C2 implant
• 3.9Nathan Lambert's written Congressional testimony on the state of open models - Chinese open-weight downloads now 2x America's, >80% of OpenRouter open-model usage
• 3.9Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war

+ 3 more in this edition

• 3.6Claude Opus 5 Helped Researchers Take Over OpenAI Staff Accounts via Chained Flaws
• 3.6Alibaba open-sources AI model that can detect cancer and nearly 150 conditions

• 4.1Everyone Says Datacenter Moratoriums Are Killing the US Buildout. We disagree
• 4.0ZCode, the GLM coding agent, silently uploads your Git history
• 4.0Open-weight models take 56% of token volume, Astra doubles Fable 5.1 spend

+ 7 more in this edition

• 4.0Exclusive: Paying for frontier AI models buys 4-month head start at 5x the cost
• 3.8Have it both ways: stay discoverable in search while disallowing AI training
• 3.5A single firm is behind OpenAI, Anthropic, and Meta hacking scandals

• 4.3Houthis used Claude Code to develop missile guidance software: Anthropic
• 4.3Apple's Siri AI Can Be Swapped Out for Claude, ChatGPT, Code Shows
• 4.3Inside ‘Project Lily’: The Humans Reading Your ChatGPT Chats

+ 1 more in this edition

• 3.8Long Live the Short King: Why 4-hi HBM Wins

• 4.3OpenAI agents attacked RubyGems back in May

• 4.5Nvidia’s Backstop Universe – Heads I Win, Tails Who Loses?
• 4.2DOD poised to move all classified AI workloads off Anthropic by October
• 4.2Anthropic: Detecting and Addressing AI Misuse by China – September 2026

+ 3 more in this edition

• 4.0Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra
• 3.9Anthropic Identifies Biased Reasoning and Recklessness as Drivers of Claude’s PyPI Attack
• 3.6Congress Pushes AI Agents Into the Audit Trail

• 4.0DeepSeek launching v4.1 flash cheaper and more capable than v4 pro
• 3.9Quoting Calif Research
• 3.9New Records Reveal Problems with Medicare’s AI Prior Authorization Experiment

+ 1 more in this edition

• 4.1Stealing AI Reasoning Traces
• 4.0Latest open artifacts (#24): Motif-3, GLM-5.3, Hy4-preview and open model licenses
• 4.0The Shared Clipboard Inside the Sandbox: Cross-Account Data Leakage in ChatGPT

+ 4 more in this edition

• 4.1TPU Inference Externalization Full Steam Ahead - InferenceX
• 3.3Nscale seeks up to $3.5B in pre-IPO financing as AI infrastructure race intensifies

• 3.6ByteDance Lands $29.6 Billion Loan to Fuel AI Advances

• 4.0Discovery of a new OpenAI agent message board
• 3.7Deepseek plans the largest known Huawei chip cluster with 160,000 processors in Inner Mongolia

• 4.7What Nvidia’s $13B acquisition of Hugging Face means for AI model choice
• 4.6GPT-6 Astra: an automated AI Engineer you can hire for <$6 an hour
• 3.6Anthropic Breaks With Peers on Massachusetts AI Safety Bill

• 4.1OpenAI Says New Model Meets Its ‘Critical’ Cybersecurity Threshold
• 3.7Three sites made 215,128 “best software” pages for AI. Perplexity cites them
• 3.7Anthropic Has Some Alignment Problems

• 3.9ChatGPT Facing Dual Regulatory Regimes Under New EU Designation
• 3.8ChatGPT Ads Hit $1 Billion Faster Than Google Ever Did
• 3.2WeChat Pay expands AI AgentPay Card to DeepSeek Harness and OpenClaw

• 3.8AI and Employment: So Far, So Good
• 3.7A Pragmatic Randomized Trial of an EHR-Integrated Generative AI Chart Summarization Tool for Ambulatory Clinicians
• 3.5OpenAI Lets Some Customers Pay Only When AI Performs

• 3.6Understanding ChatGPT Work

• 3.7Introducing Hy4 Preview

• 4.3Trump blacklisting of "woke" Anthropic deemed illegal by federal judge
• 3.5When Agents Take Over the System of Record: 40GB Nobody Typed, a Renewal Agent Built in Half a Day, and Why Our Agents Love Clay: The Agents #013

• 4.0Breaking Claude Code Opus 5 Auto Mode
• 3.7Elon Musk’s xAI used child porn to train Grok models, lawsuit says
• 3.5AI industry says Trump plans to tax chips in the “single dumbest way imaginable”

• 5.0VMs won't contain cyber-capable agents
• 4.3Inside the Warehouse Where Amazon Scans and Destroys Books for AI Training
• 4.1Anthropic Escalates Cloud Spending With $45 Billion Nscale Agreement

+ 4 more in this edition

• 5.0OpenAI Jalapeño: Better Than Nvidia Blackwell
• 3.9Alabama Probe Opens New Regulatory Front Over Containing Powerful AI Models
• 3.8Ukraine opens its massive labeled battlefield dataset to British firms in a landmark AI weapons partnership

• 3.8Nvidia to hike prices by 15%, on top of an even larger increase in July
• 3.6Nvidia reportedly eyes Perplexity at a $30B+ valuation as AI search becomes an agent business

• 4.4Poolside gets $12B reverse-execuhire to NVIDIA; founders stay for $1B, employees go for $6B, Infraco scaling to 7GW neocloud
• 4.0Are Open Models Catching Up?

• 3.3Anthropic Plans to Tweak Data Retention Rules After Enterprise Concerns
• 3.2Retailers Report AI-Driven Sales and Bigger Baskets in Q2 Earnings

• 4.1Frontier AI Application Security: Every Second Counts
• 4.1PurpleDelta's Fraudulent Employment Operations
• 4.0Microsoft Copilot reveals secret input that allowed it to be hacked

+ 1 more in this edition

• 4.1Stripe is reportedly acquiring AI startup OpenRouter for more than $7 billion
• 4.0Corporate America’s Top 1% Spend $7,400 Per Employee on AI
• 3.9Israel creates fake think tank in likely attempt to dupe AI chatbots

• 3.5Anthropic Touts 14-Fold Jump in Revenue for Would-Be Investors

• 3.6US AI Labs Cut Prices 25% in 1 Month to Fend Off Chinese Rivals
• 3.4Klaviyo’s CEO on Building at $1.5B With Agents: “Dark Factory,” Composer, and Why Every Single Employee Had to Hit L3 by June
• 3.4OpenAI’s Revenue Run Rate Tops $40 Billion as IPO Nears

• 4.1Anthropic Pursues $6 Billion Decart Deal to Cut AI Costs
• 3.9Inside the ECB’s AI Cyber Directive: What EU Banks Need to Know
• 3.8The Biggest AI Models Are Not the Biggest Threats

+ 3 more in this edition

• 4.2SpaceXAI's Grok 4.6 matches OpenAI's best model and undercuts it on price
• 3.9Your Bank’s AI Agent May Need a Permission Slip
• 3.6xAI co-founder raises $1.1B for River AI to build an open AI stack

+ 3 more in this edition

• 4.0DeepSeek overtakes Google on volume, cost per token falls 13.6%
• 3.8"But marinade" and leaked passwords are what researchers found in ChatGPT's hidden reasoning
• 3.6OpenAI introduces $125 Premium Seats for ChatGPT Business as agentic AI burns through more tokens

• 4.0Over 181,000 AI meeting recordings left wide open in note taking app

• 4.1Generative design of novel bacteriophages with genome language models [R]

• 3.5Anthropic sets Claude Code to Auto Mode by default to protect developers from bad approvals

• 4.3SpaceX 10GW in 2027 – Why It’s Real, Will Drive $300B ARR for SpaceX, and Why Microsoft Will Be the Largest Offtaker
• 3.65 Interesting Learnings from Shopify at $14B in Revenue: 34% Growth, 18% Free Cash Flow Margins, and AI Orders Up 3x

• 4.0Software Giant SAP Stops Most Travel and Hiring Because of AI’s Soaring Cost
• 3.9If You're Not Paying for the Tokens...
• 3.9Meta Model’s Hack Mirrors Previous OpenAI and Anthropic Security Breaches

+ 2 more in this edition

• 4.4Hype vs. Reality: What the Hugging Face Incident Means for AI Safety
• 4.4Anthropic and OpenAI Agents Accused of Social Engineering
• 4.3News: Microsoft Disclosures Suggest OpenAI Sales Account For Around 70% Of FY26 AI Revenue, more than 7% of FY26 Revenue

+ 4 more in this edition

• 4.6Popular npm Packages in the keyv and Cacheable Namespaces Compromised in Active Supply Chain Attack
• 4.0Appeals Court Agrees with EFF that Building a Web Browser Doesn’t Violate the CFAA
• 4.0Chinese military researchers tap US AI models to train defense systems

+ 9 more in this edition

• 4.2US company’s AI lets Ukraine’s cheap kamikaze drones track targets on their own
• 4.1The Chinese labs everyone lumps together are making four pretty different bets. I work at one of them.
• 4.0LLM Heist: Hijacking LiteLLM for Traffic Interception, Key Theft, and Tool-Call Injection

+ 7 more in this edition

• 4.0I made llama.cpp remember across restarts: 54.4s prefill -> 3.5s on a new process (free ARM box)
• 4.0Enforcement of the AI Act Starts Today
• 3.9Deepseek v4 flash - 100-150 faster t/s in prefill/pp.

+ 2 more in this edition

• 4.2Ten advances in mathematics and theoretical computer science
• 3.6German court rules AI music generator Suno violated copyrights, rejects fair use defense
• 3.5Everyone is building LLM routers, we deprecated ours

+ 1 more in this edition

• 4.4Measuring LLMs’ Ability to Perform Cryptanalysis
• 4.3Show HN: Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it
• 4.2The AI slowdown is coming

+ 11 more in this edition

• 4.4Claude Opus 5: Fable-level performance at Opus price (half Fable)
• 4.3moonshotai/Kimi-K3
• 4.2Echoverse: Deep, evolving environments for computer-use agents

+ 28 more in this edition

• 4.3How enabling two settings tripled our scores on the ARC-AGI-3 benchmark
• 4.1AI Worming through Word
• 3.8LFM2.5-Encoders for Fast Long-Context Inference on CPU

+ 1 more in this edition