OpenAI Demands State AI Training Monitoring, First Call for Development Oversight

Legislatures in California and Illinois must revisit their artificial intelligence safety frameworks to include oversight of models during development. OpenAI Global Affairs stated that current laws, including SB 53 and SB 315, are insufficient and requested that state rules require the monitoring of frontier models during training and evaluation. The company cited lessons from a security incident involving Hugging Face as evidence that models pose severe risks before they reach the release stage.

The proposed regulations would mandate the monitoring of internal model activity, a requirement OpenAI describes as a novel concept for policymakers. The developer argued that existing auditing requirements in Illinois only ensure companies follow their own safety frameworks, which may be objectively insufficient to prevent a catastrophe. This shift in position emphasizes a need for systemic oversight during the development phase rather than relying on internal company checks.