Anthropic's Mythos AI finds decade-old bugs in Firefox

Unveiled in April, Anthropic's Mythos AI model has demonstrated remarkable proficiency in identifying software vulnerabilities. The security team at Mozilla's Firefox reported that Mythos uncovered a variety of high-severity bugs, some of which had been dormant for over ten years. In April 2026, Firefox implemented 423 bug fixes, a significant increase from the 31 fixes made the previous year. The model detected sandbox vulnerabilities that human researchers had missed, despite a $20,000 bounty offered. While Mozilla employs AI to generate patches, human engineers are still essential for finalizing fixes. The overall impact on cybersecurity is still unclear; Anthropic's CEO Dario Amodei expresses optimism about defenders' prospects, whereas Mozilla's Brian Grinstead calls for caution.