Unauthorized Group Gains Access to Anthropic’s Exclusive Cyber Tool Mythos

Anthropic’s pre-release Claude Mythos Preview — an AI model described as capable of autonomously finding zero-day vulnerabilities and chaining multi-step exploits — was reportedly accessed by an unauthorized group via a third-party vendor using shared contractor accounts and API keys; Bloomberg provided screenshots and a live demo, and Anthropic is investigating while stating there is no evidence yet of impact to core systems.