Anthropic's Strongest AI Model Claude Mythos Leaked, U.S. Cybersecurity Stocks Plunge

Recently, an unpublicized blog draft of AI firm Anthropic was leaked, exposing its strongest AI model Claude Mythos under development. The official confirmed the model exists, calling it a new tier above the current top Opus, with scores far exceeding Claude Opus 4.6 in programming, academic reasoning and cybersecurity tests.

The draft noted Mythos' leading network attack-defense capabilities may pose potential security risks, so the official will adopt a cautious release strategy. The leak came from a human vulnerability due to CMS configuration errors.

Fearing super AI could become a hacker tool, U.S. cybersecurity stocks plummeted: Palo Alto Networks down 7%, CrowdStrike down 6.4%, Zscaler and Fortinet down 5.8% and 4% respectively. This isn't Anthropic's first shock—Claude Cowork caused a $285B market value drop in related sectors in February.