Anthropic's AI Finds 100+ Bugs in Firefox
In a powerful display of AI's cybersecurity potential, Anthropic's Claude model discovered over 100 bugs in the Firefox browser during a two-week test. The AI found its first exploitable vulnerability in just 20 minutes, highlighting how generative models can accelerate software security audits.
The collaboration between Anthropic and Mozilla involved the AI model Claude Opus 4.6, which analyzed Firefox's code for a two-week period in February 2026. This effort uncovered a total of 22 security vulnerabilities, with 14 of them being classified as high-severity. The AI also identified an additional 90 non-security-related bugs. The number of high-severity vulnerabilities discovered by the AI in just two weeks was significant, representing nearly one-fifth of all high-severity flaws patched in Firefox during the entirety of 2025. One report noted that the model found more high-risk bugs in that short span than are typically reported globally over a two-month period. Anthropic's team initiated the process by tasking the AI with examining Firefox's JavaScript engine, a critical area for browser security. Within the first 20 minutes, Claude identified a "Use-After-Free" vulnerability, a serious type of memory corruption flaw. This initial success prompted the broader two-week analysis. Following the discovery, Mozilla's engineers validated the findings and quickly began developing fixes. The patches for all 22 security vulnerabilities, along with the majority of the other 90 bugs, were included in the Firefox 148.0 release, which was shipped to users before any of the flaws were actively exploited.
