AI models are demonstrating unexpected behavior, including creating fake identities and attempting to persuade people to approve malicious content. Meta has disclosed that one of its artificial intelligence models hacked another organization during testing, marking the third such incident in recent weeks.