A Frontier-Model Security Audit for Open-Source Projects
Datasette ran a security audit with Claude Fable 5.1, GPT-5.6 Sol, and GPT-6 Astra. The workflow matters as much as the bugs.
Tag
#security
Hugging Face in $13B Acquisition Talks
Hugging Face is fielding bids of $13B or more, weeks after an OpenAI pre-release agent exploited its infrastructure during cyber testing.
Serving a Local Model to Your Household Without Exposing It
None of the mainstream inference servers ask for a password, and vLLM binds every interface by default. The safe pattern for household serving.
Frontier AI's 'encrypted reasoning' can be decoded for about $720
Anthropic, OpenAI, and Google ship encrypted reasoning blocks that a sibling model can bulk-decode for about $720, leaking PII and API keys.
An Agent Hacked a Gym API. Labs Are Gating Cyber Models.
In one week, an OpenClaw agent canceled a stranger's gym booking via a missing auth check, and OpenAI moved GPT-5.6-Cyber behind a partner-only Red tier.
Microsoft Copilot for Word Carries a Self-Replicating AI Worm
Håkon Måløy disclosed a self-replicating prompt injection in Word's Copilot after 144 days of coordination. Two mitigations did not close it.
Claude Cowork's SharedRoot escape: a one-prompt path to your Mac
Accomplish's Oren Yomtov chained CVE-2026-46331 to escape Claude Cowork's macOS sandbox in a single short message and reach the host filesystem.
Apple Fixed a Hide My Email Leak. Old Aliases Still Matter
Apple patched a Hide My Email flaw, but aliases created before July 7 may have exposed the real addresses they were meant to conceal.
NadMesh Targets Exposed Ollama, ComfyUI, and n8n for Cloud Keys
A Go-based botnet is scanning exposed Ollama, ComfyUI, n8n, Open WebUI, Langflow, and Gradio instances for AWS keys and Kubernetes tokens, QiAnXin XLab says.
Hugging Face's July Breach Was Run by an Autonomous Agent
Hugging Face disclosed a July 2026 breach run end-to-end by an autonomous agent. The defender was an open-weight model.
Claude's web_fetch Tool Leaked User Data Through a Honeypot
Anthropic's web_fetch tool let a prompt-injection honeypot walk Claude through user profile URLs and pull a user's name, city, and employer.
AI Security Roundup: Glasswing Finds 10,000 Zero-Days, Exploit Window Goes Negative
Anthropic's Mythos finds 10,000+ critical vulnerabilities but fewer than 1% get patched. Mandiant says exploits now arrive a week before fixes.
AI Security Roundup: One Million Exposed Services and the Teenager Problem
Intruder scanned 2 million hosts and found 1 million exposed AI services with no authentication. Plus: teenagers are using ChatGPT to hack governments, and OpenAI launches Daybreak.
Claude Mythos: The AI That Hacks Better Than Humans
Anthropic built an AI that finds zero-days autonomously. The Pentagon wants it. Anthropic said no to surveillance. Now it's a geopolitical crisis.
AI Security Roundup: OpenClaw's Nine CVEs in Four Days
OpenClaw collected nine CVEs in four days with 135,000 instances exposed. Plus: GitHub RCE, Flowise exploitation, and CrewAI trust failures.
80% of Fortune 500 Companies Have Lost Control of Their AI
Shadow AI isn't a rogue employee problem. It's a rational response to broken governance — and 90% of the security leaders tasked with stopping it are doing it themselves.
AI Security Roundup: Vercel Breached Through AI Tool, n8n's CVSS 10 Nightmare, and the Supply Chain Keeps Breaking
An AI productivity tool compromise led to Vercel customer data theft, n8n's workflow platform had an unauthenticated RCE scoring a perfect 10, and Mercor's LiteLLM-linked breach exposed training data for OpenAI and Anthropic.
AI Security Roundup: Lovable's 48-Day Open Door, MCP Remote Code Execution, and NIST Abandons 29,000 CVEs
A vibe-coding platform exposed every project's secrets through a trivial API flaw, Anthropic's MCP protocol enables remote code execution across 200,000 servers, and NIST can't keep up with AI-driven vulnerability discovery.
AI Security Roundup: Vercel Breached Through an AI Tool, North Korea Weaponized Dependabot, and Your Bot Just Auto-Merged Malware
A third-party AI tool compromise chains into Vercel's systems, North Korean hackers use Dependabot to distribute malware to 895 repos, and courts fine lawyers $145K for AI hallucinations in Q1 alone.
Your 'Safe' AI Chatbot Becomes a Liability the Moment You Give It Tools
Researchers tested five frontier LLMs as workplace agents. GPT-5.1 executed malicious instructions 75% of the time. Even the safest model failed 40%.
56% of Security Teams Can't Tell You How Fast They'd Kill Their AI
ISACA surveyed 3,400 security professionals. Most don't know how quickly they could shut down an AI system during an incident. One in five doesn't know who's responsible.
AI Security Roundup: Mercor's 4TB Breach, n8n and Langflow RCEs, and the Week AI Workflow Tools Became Attack Surfaces
A supply chain attack exposes 40,000 AI contractors, three major workflow platforms get critical RCE flaws, and Microsoft patches 167 vulnerabilities as AI-driven discovery triples submission rates.
Your AI's Safety Filter Fails 97% of the Time Under Fuzzing
Palo Alto's Unit 42 tested LLM guardrails with genetic-algorithm prompt fuzzing. Content filters missed up to 99 out of 100 attacks.
One Line of Code Jailbreaks 11 AI Models — Including the 'Safe' Ones
Trend Micro confirms the sockpuppeting attack bypasses ChatGPT, Claude, and Gemini using a basic API feature. Some providers have patched it. Most haven't.