“OpenAI's strategy for developing safe AI systems is an iterative process of deploying models while stakes are relatively low to gather real-world feedback and identify issues.”