Mozilla Fixes 423 Firefox Vulnerabilities with AI

Mozilla has reported fixing 423 security vulnerabilities in its Firefox browser during April 2026, with Anthropic's experimental AI model, Claude Mythos Preview, playing a central role in identifying 271 previously unknown bugs in Firefox 150 alone. The effort marks a sharp increase from 76 fixes in March 2026 and 31 in April 2025, according to a detailed post on the Mozilla Hacks blog by three Firefox developers.

The breakthroughs stemmed from Mozilla's custom agentic pipeline, which integrated Mythos with the browser's fuzzing infrastructure. This harness provided the AI with targeted code files, read-write access, and sanitizer builds that crash on memory errors. The model generated reproducible test cases, which a second AI instance verified, drastically reducing false positives. "In terms of the bugs coming out on the other side, there are almost no false positives," said Mozilla Distinguished Engineer Brian Grinstead.

Of the total fixes, 271 came directly from Mythos scans of Firefox 150, while about one-third of the remaining 111 internal discoveries also traced to the pipeline using Mythos or similar models. Traditional fuzzing and 41 external reports accounted for the...