Anthropic’s tightly controlled rollout of Claude Mythos has taken an unexpected turn. Despite weeks of assurances that the AI model is too capable at cybersecurity to be released publicly, it appears that Mythos has fallen into the wrong hands.

According to Bloomberg, a small group of unauthorized users has had access to Mythos since the day Anthropic announced plans to offer it to a select group of companies for testing. This breach is particularly embarrassing for Anthropic, which has built its brand on prioritizing AI safety while touting the cybersecurity capabilities of its models.

Anthropic has confirmed that it is investigating the incident, but the fact that the model was compromised despite the company’s claims of robust security measures raises questions about the effectiveness of its safety protocols.

Photo by Mariano Di Luch on Pexels
Photos provided by Pexels