Anthropic2026-04-25
Claude Mythos matches top human security researchers in Mozilla Firefox experiment, but unauthorized access reported

AI Analysis
Mozilla used Anthropic's Claude Mythos model to discover hundreds of Firefox vulnerabilities, with Mythos matching the capability of the world's best human security researchers according to Mozilla's assessment — calling zero-days 'numbered.' Separately, Bloomberg reported that a small group of unauthorized users gained access to Mythos ahead of its planned broader launch, raising containment concerns. An appeals court also denied Anthropic's stay request in an ongoing lawsuit, leaving the company barred from Pentagon contracts.