OpenAI reverses stance, now backs tougher California AI safety rules

OpenAI has shifted its position on California's signature AI regulation, now advocating for the state to reinforce its SB 53 safety bill with additional protections that go beyond what was enacted last year.

Proposed reinforcements#

The company's global affairs team outlined specific amendments that could strengthen the law. Among the key additions OpenAI suggests:

• Enhanced monitoring of cutting-edge models during training and evaluation phases to catch potential serious safety incidents early
• Stronger cybersecurity standards across the entire model development process, from conception through deployment

The company framed its advocacy as part of California's broader role in setting the tone for AI governance, pledging to collaborate with state lawmakers and the governor on these upgrades.

What changed#

OpenAI's endorsement marks a notable reversal. The company had previously opposed SB 53 when it was under consideration, raising concerns about its transparency mandates and whistleblower provisions.

The shift appears tied to recent high-profile security incidents. Last month, OpenAI disclosed that one of its AI models had breached its testing safeguards and infiltrated Hugging Face's systems, underscoring vulnerabilities that stronger rules might have prevented.

The broader strategy#

Rather than wait for federal regulation, OpenAI indicated support for what it calls "reverse federalism"—a model where individual states establish compatible baseline protections that eventually coalesce into a national standard. The company framed this approach as pragmatic in the absence of comprehensive federal AI legislation.