Discord group says it accessed Claude Mythos by guessing location
6 hours ago
- #AI Security
- #Cybersecurity
- #Data Breach
- An anonymous Discord group claims to have accessed Claude Mythos Preview by guessing its online location based on Anthropic's naming conventions.
- The group included a member with privileged access as a contractor for Anthropic, aiding in breaching security.
- Claude Mythos is described by Anthropic as capable of identifying and exploiting zero-day vulnerabilities, posing a significant cybersecurity threat.
- Anthropic is investigating the claim, but there is no evidence of broader unauthorized access at this time.
- The group asserts they used the model for non-malicious tasks like building websites, but also claim access to other unreleased Anthropic models.