And 14 of the security vulnerabilities detected were classified as high risk.
Anthropic, in collaboration with Mozilla, identified 22 security flaws in the Firefox browser during a two-week test, with 14 of the vulnerabilities classified as serious.
The discoveries were made using the AI model Claude Opus 4.6. It began by analyzing the Firefox JavaScript engine and then moved on to other parts of the code base. In total, the model examined nearly 6,000 C files and generated 112 error reports.
Most of the vulnerabilities had already been fixed in Firefox 148, which was released in February. However, some of the fixes are not expected until the next version of the browser is released.
The test also showed that Claude is significantly better at finding security flaws than at writing code to exploit them. Despite spending around $4,000 in API credits, the team only managed to exploit two of the bugs.
