Mozilla used early access to Anthropic's Mythos Preview to identify and patch 271 vulnerabilities in Firefox 150, shipped this week.
Mozilla's Firefox team partnered with Anthropic to get early access to Mythos Preview, a new AI model with advanced cybersecurity capabilities. The tool identified 271 vulnerabilities now patched in Firefox 150, released this week. Firefox CTO Bobby Holley described Mythos as covering 'the full space of vulnerability-inducing bugs' — a step-change from traditional fuzzing and manual review. Anthropic and OpenAI have both restricted these models to limited private releases due to the dual-use risk of putting them in attackers' hands.
Mythos Preview reportedly covers the 'full space of vulnerability-inducing bugs' — not just known patterns but novel classes that traditional fuzzing and static analysis miss. This isn't incremental; Mozilla's team described it as a firehose requiring organizational adaptation to keep up with the output volume. The risk mirror is real: the same capability will reach attackers when these models go public.
Run Semgrep or CodeQL on your most security-critical module this week to establish a baseline bug count — then you'll have a real number to compare against when Mythos-class tools become available to your team.
Go to claude.ai and open a new conversation
Tags