Skip to content
Live
Trending storyPast story

Anthropic internal model discovers real software vulnerabilities

1 reports1 outlets21 days agoUpdate

The story so far

AI overview

Anthropic claims to have discovered critical vulnerabilities in widely used software such as Chrome, Safari, Android, Windows, and Linux using an unreleased internal research model, linking them into five attack chains that can remotely execute code or cause other severe impacts. Anthropic states that each vulnerability has been reported to the affected vendors following the coordinated vulnerability disclosure process, with vendors given 90 days to fix them, and then 45 days before discussing content beyond public announcements. Subsequent reports indicate that all related vulnerabilities have been reported to vendors and fixed: For Chrome, Anthropic's report generated 62 severe or high-risk CVEs, including 13 sandbox escape vulnerabilities and 21 vulnerabilities that can control web rendering processes, with fixes released alongside Chrome 156; Safari's vulnerability chain can bypass the same-origin policy, with Apple confirming and releasing fixes; Android attacks require physical contact with the device, allowing personal data extraction such as photos while the Pixel 9 is locked, with Google stating it has addressed this. Anthropic also claims to have recently expanded the Cyber Verification Program, providing advanced network capabilities, reduced interception classifiers, and models such as Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1 to qualified security professionals. These statements all come from Anthropic itself.

AI-written from the reports · Updated 21 days ago

Coverage timeline

Follow the coverage to see every side of the story.

27 Oct
  1. Anthropic ResearchFeatured
    Discovering and exploiting vulnerabilities in critical, real-world software with Claude

    Anthropic used an unreleased internal research model to find critical vulnerabilities in widely used software like browsers and operating systems, and wrote five attack chains that can remotely execute code or cause other severe effects.