Anthropic Finds 22 Firefox Vulnerabilities Using Claude Opus 4.6 AI Model

Anthropic used its Claude Opus 4.6 LLM in partnership with Mozilla to scan Firefox source code and found 22 vulnerabilities (14 high severity); Mozilla addressed most in Firefox 148. The report highlights a high-impact JIT miscompilation (CVE-2026-2796, CVSS 9.8) and notes that the LLM was able to produce working crude exploits in two controlled lab cases, though no evidence of real-world exploitation was reported and many issues were patched.