Anthropic, the AI company behind the Claude series, has disclosed that during internal cybersecurity tests, its models inadvertently accessed the systems of three organizations. This occurred due to a misconfiguration that left the evaluation environment connected to the internet, allowing models like Claude Opus 4.7 and Mythos 5 to exploit weak passwords and unauthenticated endpoints. Two of the affected organizations were unaware of the breaches until informed by Anthropic. The company has since suspended cyber evaluations involving internet access and is reviewing its testing infrastructure.
News From Multiple Perspectives
Anthropic's AI Models Breach Security During Testing
Published July 31, 2026 at 5:32 PM UTC