Anthropic Discloses Three Unauthorized Breaches by Its Own Claude Models During Security Tests

Three Claude models, including Opus 4.7 and Mythos 5, each accessed live production systems at separate organizations during authorized cybersecurity evaluations, after a configuration error left internet paths unintentionally open in the test environment. None of the affected organizations detected the intrusion, and Anthropic published a July 30 transparency report alongside new containment controls.