Anthropic recently shared details of a collaboration with Mozilla researchers where the AI model Claude Opus 4.6 discovered 22 vulnerabilities in Firefox over two weeks. According to Anthropic, Mozilla assigned 14 of these as high-severity vulnerabilities, representing nearly a fifth of all high-severity Firefox vulnerabilities remediated in 2025. After an initial discovery in the JavaScript engine, the effort expanded across nearly 6,000 C++ files, resulting in 112 unique reports submitted to Mozilla and fixes shipped to hundreds of millions of users in Firefox 148.0. To measure the upper limits of Claude's cybersecurity capabilities, Anthropic also tested whether the model could create exploits for the discovered bugs. Researchers found that Claude was significantly better at finding vulnerabilities than at exploiting them, successfully generating a working exploit in only two out of several hundred test cases using an environment with security features removed. Anthropic noted that these findings underscore the importance of accelerating the find-and-fix process for defenders and offer a model for how AI-enabled researchers and software maintainers can collaborate.
Partnering with Mozilla to improve Firefox’s security
Anthropic announced a collaboration with Mozilla in which the AI model Claude Opus 4.6 discovered 22 vulnerabilities in Firefox over two weeks. Mozilla assigned 14 of these as high-severity vulnerabilities, which were shipped in Firefox 148.0. The partnership highlights both the accelerating speed of AI-driven vulnerability detection and the ongoing challenges surrounding automated exploit development.
Why it matters
This partnership demonstrates that advanced AI models can independently uncover critical software vulnerabilities at accelerated speeds, offering powerful new tools for both software defenders and researchers.
Topic dossiers Anthropic & Claude