Anthropic confirmed a January 2026 security incident involving Claude Opus 4.6, marking its fourth public disclosure. Automated screening initially missed the breach during a cybersecurity assessment where a configuration error connected the model to the open internet. Discovered during an August review, Anthropic expanded its investigation to 481 million logs, re-identifying the four incidents through a second deep review of 9.2 million high-risk entries without finding additional severe cases.