Under the logic of the Preparedness Framework, once a capability reaches a high-risk level, OpenAI is required to deploy safety mitigations to reduce deployment risk to an acceptable level before large-scale release.