Anthropic's Mythos AI Model Unauthorized Access

Anthropic's new Anthropic — Mythos (AI model), an advanced AI system with significant cybersecurity capabilities, has been accessed by a small group of unauthorized users. The breach, first reported by Bloomberg L.P., occurred on the same day Anthropic announced limited testing of the model under its Project Glasswing initiative. The unauthorized access was allegedly linked to a third-party contractor environment and involved users coordinating through a private Discord channel. These users reportedly leveraged knowledge of Anthropic's system patterns and potentially information from a separate data breach at Mercor to gain entry. While Anthropic has stated there is no evidence the breach extended beyond a vendor environment or impacted its core systems, the incident raises serious concerns about the security of advanced AI models. Anthropic — Mythos (AI model) is capable of detecting and exploiting thousands of critical flaws, including zero-day vulnerabilities, and has been made available to approved partners like Amazon (company), Apple Inc., Cisco, Nvidia, Alphabet Inc., and Microsoft. The event underscores the broader challenge for AI companies in controlling access and usage of powerful AI systems.