Claude Mythos Helped Firefox Fix More Security Bugs in April Than Prior 15 Months Combined

Mozilla's blog documents that deploying Claude Mythos Preview enabled the Firefox security team to fix more bugs in April 2026 than in the previous 15 months combined. Anthropic held vulnerability details for approximately one month before release to give defenders patching time; information is now being disclosed via the Mozilla blog. Multiple independent commentators — including Gary Marcus and Ethan Mollick — confirmed the capability as real but bounded: not an indiscriminate exploit generator, but a measurable acceleration of legitimate security research.

Why It Matters

This is the most concrete quantified demonstration of frontier AI in production security work. It frames the Anthropic-vs-OpenAI security competition (GPT-5.5-Cyber vs Mythos) around operational outcomes, not just benchmark claims.