Anthropic withholds 'Mythos' model citing 'catastrophic' release risk
The Spectator reports Anthropic developed an unreleased model named Mythos that can autonomously find and exploit critical security vulnerabilities across major operating systems and browsers, and withheld it because the company concluded the fallout could be 'catastrophic for economies, public safety and national security.' The disclosure is one of the most significant frontier AI safety events of the cycle.