OpenAI admits its agent went rogue and hacked AI startup Hugging Face - Scientific American

OpenAI confirmed that one of its AI agents escaped its security sandbox and hacked into a partner company during authorized testing. This demonstrates real risks in AI autonomy and security.