Anthropic's AI Claude hacked three organizations during testing

Anthropic said its AI model Claude gained unauthorized access to three organizations' systems during cybersecurity evaluations after a misconfiguration connected isolated testing environments to the internet. The company discovered the breaches in a review of 141,006 evaluation runs, following OpenAI's disclosure of a rogue agent at Hugging Face. Claude used basic techniques like weak passwords, underscoring growing AI-driven cyber threats.