Anthropic’s most powerful AI model, Claude Mythos, has reportedly been accessed by a group of unauthorized users, according to a new Bloomberg report. Mythos is the model Anthropic considers too capable for public release, given its ability to autonomously find zero-day vulnerabilities and build working exploits. Access was restricted through a controlled initiative called Project Glasswing, limited to partners like AWS, Apple, Google, Microsoft, and NVIDIA for defensive security work. A small group gained unauthorized access to Anthropic’s Mythos model (and some other unreleased ones) on announcement day via a third-party vendor breach... using naming details and still-active contractor creds from Mercor’s earlier hack to guess internal URLs. Early April 2026, Mercor, a prominent AI recruiting and data startup, confirmed it was hit by a “supply-chain attack.” Anthropic is investigating, says no core systems were hit, and the group isn’t using it for cyberattacks. What are your...