Claude AI in Real Vulnerability Hunting: Extracting Photos from a Locked Android Phone and Gaining Root on Linux in 53 Minutes

Anthropic says an unreleased internal Claude research model found vulnerabilities across Chrome, Safari, Android, Windows and Linux, chaining them into five working exploit chains. Its report to Google produced 62 severe or high-severity CVEs, including 13 sandbox escapes, and a Linux attack gained a root shell in 53 minutes; all flaws were reported through coordinated disclosure and fixed. It matters because it shows frontier models can now autonomously locate and exploit real vulnerabilities, raising the bar for defensive testing as well as for responsible disclosure.

• IT之家2026-10-28 22:35