Security Desk
Security AI news and analysis
AI vulnerabilities, cyber incidents, model abuse, and practical defenses for securing AI systems.
14 recent briefs · newest first
Elastic Security Labs · Aug 4, 2026, 2:24 PM EDT
Agents vs. agents: how we triage HackerOne reports for $2 each, 85% as well as a human
Elastic now triages its surging, largely AI-generated bug bounty reports with its own AI for about $2 each, replacing 30–60 minutes of senior engineer time per report.
Socket · Aug 4, 2026, 8:24 AM EDT
Popular npm Packages in the keyv and Cacheable Namespaces Compromised in Active Supply Chain Attack
Any project that installed keyv or cacheable-family packages since August 4, 2026 may have leaked cloud and CI credentials now being used to trojanize more npm packages.
Cisco Talos · Aug 4, 2026, 7:11 AM EDT
“Keep going, bro. You’ve got this!” A data-driven look at how adversaries are weaponizing AI
Talos's analysis of criminals' abandoned chat logs shows AI guardrails rarely stop misuse — and gives defenders a new forensic trail.
Embrace The Red · Aug 3, 2026, 1:12 PM EDT
LLM Heist: Hijacking LiteLLM for Traffic Interception, Key Theft, and Tool-Call Injection
A compromised LiteLLM gateway hands attackers every backend LLM provider key plus the ability to reroute, read, and alter model traffic and tool calls.
The Decoder · Aug 2, 2026, 9:12 AM EDT
A real macOS flaw worth $200K went unreported because Apple's bug bounty inbox was full of AI slop
AI-generated junk reports have so clogged Apple's bug bounty queue that a real macOS flaw worth up to $200,000 initially went unreported.
The Hacker News · Jul 31, 2026, 1:32 PM EDT
Chinese Hacker Commands DeepSeek via Telegram to Launch Autonomous Attacks
Demonstrates that threat actors are already operationalizing open-source agentic frameworks with frontier models for fully autonomous cyberattacks.
Schneier on Security · Jul 31, 2026, 1:22 PM EDT
Measuring LLMs’ Ability to Perform Cryptanalysis
Frontier models are now discovering novel mathematical breaks in NIST cryptographic candidates, signaling an imminent shift in how security primitives are evaluated and deployed.
Embrace The Red · Jul 31, 2026, 1:14 PM EDT
Escaping Linux Sandboxes via PipeWire (CVE-2026-5674)
Details a critical Linux sandbox escape via PipeWire that compromises the isolation of containerized AI agents, requiring immediate patching for secure deployments.
Trail of Bits · Jul 31, 2026, 1:06 PM EDT
How we use /goal to find bugs in Patch the Planet
Trail of Bits demonstrates that letting Codex write its own goal prompts significantly improves autonomous bug hunting in critical open-source codebases.
Ars Technica AI · Jul 30, 2026, 6:15 PM EDT
We now have a better understanding how OpenAI hacked into Hugging Face
The disclosure of the specific JFrog Artifactory zero-day used by OpenAI's agents provides a critical patch-and-monitor priority for teams deploying autonomous agents in enterprise environments.
Ars Technica AI · Jul 30, 2026, 5:22 PM EDT
Anthropic is finding bugs faster than Microsoft can fix them
AI-driven vulnerability discovery is now outpacing human remediation cycles, forcing security teams to rethink patch management and threat modeling.
NVIDIA Developer Blog · Jul 30, 2026, 5:12 PM EDT
How to Self-Host a Validated AI Coding Assistant with NVIDIA NeMo Guardrails
Provides a concrete architectural pattern for deploying secure, compliant, and auditable AI coding assistants in regulated enterprise environments.
Simon Willison · Jul 30, 2026, 12:02 PM EDT
An Inside Look at the Relay Market Powering Token Resellers and Fraud
Exposed LLM endpoints are actively targeted by sophisticated relay networks for token arbitrage and model distillation, making strict API spend caps mandatory for public deployments.
Simon Willison · Jul 29, 2026, 10:05 PM EDT
AI Worming through Word
Enterprise teams using Copilot for Word must restrict document ingestion from untrusted sources to prevent self-replicating prompt injection worms.