AI Identifies Security Vulnerabilities in Firefox

Anthropic’s advanced artificial intelligence model, Claude Opus 4.6, has discovered several security vulnerabilities in the Mozilla Firefox browser. The San Francisco–based company collaborated with Mozilla to evaluate how effectively the AI could identify bugs in real-world software environments. Within just two weeks, the model detected 22 vulnerabilities, including 14 that were considered high severity security risks.

To test the system’s capabilities, researchers first trained the model using a dataset containing older Firefox security issues. The AI successfully reproduced many previously known vulnerabilities and was then assigned the task of locating new ones in the latest version of the browser. Initially, the research focused on Firefox’s JavaScript engine before expanding the analysis to other components of the browser.

During the experiment, the AI reviewed nearly 6,000 C++ files and generated 112 unique security reports. After verification by researchers, these findings were shared with Mozilla, which confirmed 22 vulnerabilities. Most of these issues were fixed in the Firefox 148 update, while the remaining problems are expected to be resolved in future updates. Anthropic stated that about $4,000 in API credits were used for the experiment, and Mozilla has now begun using the AI internally to strengthen browser security.