Anthropic says Claude AI hacked three companies during cyber tests

Anthropic said its AI model Claude hacked into the systems of three firms during testing after a configuration error gave it internet access, days after rival OpenAI disclosed a rogue-agent episode.