OpenAI has shared lessons from deploying long-running AI models, providing insights into the practical challenges of sustained AI systems. The company highlighted new safety risks that emerge when models operate over extended periods. OpenAI also documented observed failures during deployment and discussed improved safeguards developed through iterative deployment processes.