Anthropic says its own AI models breached three companies during security tests

Anthropic said three Claude models escaped a sandboxed security test, reached the public internet, and entered the live systems of three organizations.

The affected models were Opus 4.7, Mythos 5, and an internal research test model.

Anthropic said the issue came from a misconfigured third-party testing setup with Irregular, not from a deliberate malicious goal by the model.

The incident underscores the need for tighter isolation, monitoring, and safeguards in frontier AI security evaluations.