Mozilla Unveils 271 AI‑Detected Security Fixes for Firefox 150

Firefox has shipped a large batch of security fixes after an AI‑driven audit uncovered 271 previously unknown bugs in the 150 release.

The audit was performed with Claude Mythos Preview and earlier with Claude Opus 4.6, using a custom harness that generates reproducible test cases and integrates with the existing fuzzing infrastructure. The bugs span a range of categories, including sandbox escapes, use‑after‑free, race conditions, and buffer over‑reads, and are identified by IDs such as 2024918, 2024437, and 2021894. Fixes were rolled out in Firefox 149.0.2, 150.0.1, and 150.0.2, with the 150 release receiving the bulk of the patches. The pipeline’s modular design allows the team to swap in newer models as they become available, improving detection rates and proof‑of‑concept generation. The effort involved over 100 contributors, from patch writers to triage engineers, and demonstrates the viability of LLM‑assisted security hardening at scale. Mozilla’s decision to publish a small sample of the reports publicly signals a shift toward transparency and encourages other projects to adopt similar AI‑driven workflows. The work underscores that AI models can now find complex, multi‑process bugs that traditional fuzzers miss.