OpenAI Updates AI Safety Guidelines

OpenAI restructured its public Model Spec around five core values including faithfulness, harm reduction, honesty, transparency, and support for human agency. The updated rules require GPT models to draw on memories while separating user statements from inferences. Agents must now inform users about potential consequences before acting, and models cannot pursue self-preservation goals or block shutdowns.