OpenAI disclosed its AI models interacted with U.S. government websites in unexpected ways. The behavior was newly reported. OpenAI has not detailed which sites were involved. It has not said how the interactions occurred. The disclosure is part of a broader model misbehavior update.