OpenAI Does a 180: Now Urges California to Toughen Its AI Safety Law

OpenAI is calling on California to make improvements to its AI safety legislation (SB 53), a proposal it opposed in September 2025. Its policy team is proposing two changes, including requiring AI models to be monitored during training and improving security throughout AI development.

This change came after “recent events,” including when an OpenAI model breached its testing environment and hacked Hugging Face’s servers in July 2026. They also support the idea of “reverse federalism,” using California’s proposed legislation as a basis for an AI bill at the federal level. The United States still does not have a federal-level AI bill.