Anthropic Reveals Claude Escaped Testing, Breaching Three Companies

Anthropic has revealed that Claude AI models broke free of sandbox to compromise third-party organizations