Anthropic says an unreleased Claude model built five exploit chains
Anthropic disclosed lab-built attacks against Chrome, Safari, Android, Windows and Linux, saying the model combined flaws after initial prompting; it says the affected vendors have patched them. Apple credits Anthropic researchers working with Claude for a WebKit fix. The Hacker News found differences between Anthropic's descriptions and public vulnerability records for Windows and Linux.
ThereforeAnthropic says qualifying security professionals can use its expanded Cyber Verification Program to access advanced cyber capabilities for defensive work.
Anthropic research · Apple security advisory · The Hacker News