Mozilla Used Anthropic’s Mythos to Find and Fix 271 Bugs in Firefox

Mozilla announced on Tuesday that its Firefox 150 browser release includes security fixes for 271 vulnerabilities identified using early access to Anthropic's Mythos Preview model. Mozilla's team fed targeted prompts into Mythos alongside their existing code-analysis pipeline; the model flagged a range of memory-safety bugs that automated static-analysis and fuzzing tooling had missed. Mozilla plans to continue the collaboration and share methodology findings with the open-source security community.

Mythos, launched by Anthropic in early 2026 as a general-purpose AI 'advisor,' is designed for enterprises seeking to deploy AI in safety-critical workflows. Anthropic selected 'a small number of trusted partners' for the Preview — Mozilla being among them. The project was framed as a test of Mythos's capacity for security and code review. Mozilla said Mythos found issues 'including critical- and high-severity vulnerabilities that could lead to code execution or other adverse outcomes for Firefox users.' It declined to identify specific software components until a full advisory is published.

The partnership comes as Anthropic is facing scrutiny over a reported security breach in which unauthorized users accessed Mythos prior to the Mozilla agreement. It is unrelated to Anthropic's corporate or government regulatory legal disputes, including ongoing investigations from Canadian data authorities. Mozilla's adoption of Mythos places it among a widening pool of groups working with big labs to incorporate AI tooling into non-model software delivery; observers said this suggests a movement away from traditional fuzzers for professional code quality assurance. Financial terms of the partnership were not disclosed.