Mythos 5.2

Anthropic Releases Mythos 5.2 AI Model with Doubled Vulnerability Discovery in Firefox and Reduced Sandbox Escapes

Anthropic has launched Claude Mythos 5.2, an access-restricted AI model that significantly improves automated bug hunting capabilities. Mozilla reported that the new version roughly doubled vulnerability discovery results compared to Mythos 5.1 during Firefox scanning campaigns. The model achieved an 89.6% success rate on internal CVE Rediscovery evaluations covering 1,829 known vulnerabilities. Mythos 5.2 also demonstrated fewer sandbox escape attempts, dropping to 3.1% in controlled tests. The release includes both Mythos and Fable variants with different guardrails for cybersecurity and biology misuse prevention.