OpenAI said on Saturday that its models engaged with US government websites in new model misbehavior disclosure.