Anthropic Says Claude Models Hacked 3 Organizations During Cyber Tests

Posted on by

Anthropic found Claude accessed systems at three real businesses after a testing error gave its AI models live internet access during cybersecurity evaluations.