Post
HIGH

Claude Mythos Preview Found 271 Firefox Vulnerabilities in Anthropic-Mozilla Collaboration

· ai-safety · vulnerability · anthropic · llm

Firefox 150 ships with fixes for 271 vulnerabilities discovered by Anthropic’s Claude Mythos Preview as part of a collaboration with Mozilla. Firefox engineering lead Bobby Holley described the process as requiring the team to “reprioritize everything” to handle the volume of findings, while calling the outcome a hopeful demonstration of AI-assisted security review at scale.

The scale of the result — 271 vulnerabilities from a single AI-assisted evaluation of one codebase — signals a step-change in the economics and throughput of automated vulnerability discovery. Security teams should follow Mythos availability announcements closely; the Firefox collaboration provides early evidence of the model’s real-world audit capability and sets expectations for what similar engagements may produce for other large codebases.