OpenAI announced on September 9 its support for establishing a mandatory national AI safety framework in the United States. The company proposes that leading AI laboratories should be subject to standardized testing, independent evaluations, cybersecurity protocols, and mandatory reporting of major incidents based on model capabilities and risk levels.
The company officially endorsed four California AI bills, shifting its previous stance due to rapid advancements in model capabilities. OpenAI emphasized that these regulatory obligations should target well-resourced organizations rather than startups or individual researchers to maintain market competition and prevent research from moving overseas.
The supported California legislation includes SB 813, which creates a framework for independent AI verification, and AB 1405, which establishes a registry and professional standards for AI auditing firms. Governor Gavin Newsom signed both of these bills into law on September 9.
Additionally, OpenAI supports SB 1119, focusing on safety measures for minors using companion chatbots, and AB 1864, which mandates federal screening standards for gene synthesis services to mitigate biological threats. These two bills were awaiting the governor's signature at the time of the announcement.
OpenAI advocates for mechanisms that would pause model development if safety thresholds are not met. The company also clarified that front-end safety regulations should not lead to broad restrictions on open-weight models, as this could consolidate the dominance of large corporations.
