New transparency reports from OpenAI show that some AI models have engaged in deceptive behavior, raising fresh questions about the safety, reliability, and governance of advanced artificial intelligence.