Anthropic's internal model found and exploited critical bugs in Chrome, Safari, Android and more
emmanuelvivier · x · 2026-10-30
Anthropic's Frontier Red Team reveals that over recent months an unreleased internal research model discovered critical vulnerabilities in the world's most widely used software, with five full attack chains published — all enabling remote code execution or other critical impact with little to no user interaction. All were reported under coordinated disclosure (90-day patch window plus 45-day embargo).
- Chrome one-click RCE: Claude found a V8 heap corruption bug, then a second bug to escape V8's memory sandbox and defeat ASLR, building a ROP chain for full control. The PoC launches Calculator on a stock macOS Chrome within seconds of page load — no user interaction.
- Additional chains target Safari, Android, Windows, and Linux, spanning open-source and proprietary software, showing the model can build sophisticated multi-step exploits.
- These products affect billions of users, making them a bellwether for AI offensive capability — offensive AI security is no longer theoretical.
More from Models
- Rumors swirl of DeepSeek Cyber (personal/enterprise, KYC required) and DeepSeek Nano — gnukeith · 2026-10-30
- Insider view: OpenAI's internal-external gap widens while Anthropic ships faster and safer — haider1 · 2026-10-30
- Claude quotes the public-domain KJV Bible but not the copyrighted NIV, over copyright limits — StewartalsopIII · 2026-10-30
- Claude feels like a leap forward while Google's AI Mode still feels like search with a bolt-on — RetoricEuphoric · 2026-10-30
- Mistral touts new large-4 update that ranks 2nd among open-source models on coding benchmark — MistralAI · 2026-10-30
- Run Kimi K3 at home: tinygrad hits 40 tok/s interactive for about $30k — StephanSturges · 2026-10-30