Anthropic rolled out Mythos to a handful of corporate testers on April 7. On that very day, a private online forum managed to slip past security and grab access. The group didn’t hack in later—they signed up with stolen or leaked credentials right alongside approved partners. Since then, they’ve been running queries through Mythos almost daily. A person with direct knowledge shared screenshots and even staged a live demo. They weren’t poking at firewalls or hunting vulnerabilities; their use cases remain unclear. Anthropic hasn’t said how many people are in the forum or what they’re building with the model. Bloomberg Intelligence analyst Matt Bloxham noted on camera that this breach underscores how sensitive cutting-edge AI remains, even when released under tight controls. Anthropic hasn’t publicly addressed the specific lapse beyond telling customers it’s “investigating.” Internally, the company is scrambling to rotate keys, tighten API gates and track down the leaked login tokens. Anthropic faces fresh pressure after rolling out Mythos under strict NDAs and usage caps. Investors, partners and regulators now all have questions: How did these credentials escape? What data might have been exposed? And will this incident slow down future pilots with big tech players like Apple and Amazon?