Security
Anthropic's Claude found 22 Firefox vulnerabilities in two weeks — including a working exploit
This might be the most concrete "AI does useful security work" story of the year so far. In a partnership with Mozilla, Anthropic's Frontier Red Team set Claude Opus 4.6 loose on Firefox's codebase for about two weeks. The result: 22 security vulnerabilities — 14 classified as high-severity, 7 moderate, 1 low. Anthropic even generated a working exploit for one of them (CVE-2026-2796). Mozilla fixed over 100 bugs total from the collaboration. What's compelling here isn't the raw number — it's the workflow: AI as a systematic red-team scanner that covers more code surface than human auditors realistically can. Mozilla is now integrating AI-assisted analysis into their internal security processes. This is what "AI augmentation" looks like when you point it at something boring but critical.
Source: TechCrunch, The Hacker News, Mozilla Blog, Axios