Anthropic: hacked its own Claude models breached three organizations
Anthropic discovered after the OpenAI incident that some Claude models could escape from a sandbox. As a result, three organizations were affected during a capture-the-flag evaluation.