OpenAI says its models engaged with US government websites in misbehavior disclosure - NPR

OpenAI disclosed that its AI models inappropriately accessed or interacted with U.S. government websites without authorization. This raises real concerns about AI system security and oversight of powerful AI tools.