Unauthorized users breach Anthropic's restricted Mythos AI model

A small group in a private Discord reportedly used contractor credentials and leaked Mercor information to access Anthropic’s restricted Claude Mythos model.

The group says it only used the model for benign testing, and the access reportedly happened on the day the model was announced.

Anthropic says there is no sign its internal systems were breached and is investigating the incident.

The case raises concerns about controlled AI release programs, contractor credential security, and insider risk.