OpenAI’s work on deliberative alignment makes an important point: safety may improve when systems can reason through rules rather than only follow patterns.