Anthropic's Leaked Cyber AI Raises Alarm Bells

Anthropic unveiled Claude Mythos in April 2026, a restricted AI model capable of identifying zero-day vulnerabilities across major operating systems and browsers, including one flaw dormant for 27 years. The UK AI Security Institute confirmed it completed a 32-step simulated cyberattack, a first for any AI model. Access was limited to roughly 40 partner firms through Project Glasswing. Despite tight controls, a small group of unauthorized users reportedly obtained the model via a private online forum. Regulators in the US and UK have taken notice, with Mythos added to financial stability agendas. Anthropic has said little about the breach.