Anthropic's AI model Claude Opus 4.6 discovered 22 vulnerabilities in Mozilla's Firefox browser during a two-week security audit, with 14 classified as high-risk vulnerabilities. The team focused on the complex and thoroughly tested Firefox codebase, spending $4000 worth of API credits to generate proof-of-concept exploit code, but ultimately only successfully exploited two vulnerabilities. Most of the vulnerabilities have been fixed in Firefox 148, with remaining fixes scheduled for the next release. This highlights the enormous potential of AI in the open-source security field, despite the ongoing challenges in creating exploit code.