Anthropic red team: unreleased Claude model built working exploit chains for Chrome, Safari, Android, Windows and Linux
Anthropic's Frontier Red Team says an internal research model found bugs and chained them into five proof-of-concept exploits in lab settings. They include a one-click Chrome chain and a zero-click Linux root exploit. Every flaw was disclosed to the vendors and patched first. Reporters note that some CVE records don't clearly match Anthropic's descriptions.
Why it matters
Shows that frontier models can now chain bugs into full exploits, so defenders should expect patch windows to shrink and AI-assisted attackers to arrive.
Source: Read the full story on Anthropic