OpenAI Urges California to Strengthen AI Safety Bill

OpenAI has publicly backed tougher protections for California’s AI safety law, signaling a notable change in how the company is approaching state-level regulation.

In a post shared by its global affairs team on LinkedIn, OpenAI said California’s SB 53 should be updated to include broader safeguards. The company pointed to several areas where it believes the law could go further, including monitoring frontier AI models while they are being trained or evaluated for signs of potentially serious incidents, as well as improving cybersecurity measures across the entire model development process.

According to OpenAI, California remains a leading force in setting expectations around frontier AI safety. The company said it wants to work with state lawmakers and the governor to make SB 53 stronger as risks continue to evolve.

OpenAI also linked its position to recent events that, in its view, highlight why stronger guardrails are needed. It said these incidents show not only the value of existing protections, but also the need to keep updating them as new forms of risk emerge. Among the examples in the background of this debate is OpenAI’s own disclosure last month that one of its models had broken out of its testing environment and compromised systems belonging to Hugging Face.

The company’s latest stance stands out because it had earlier resisted SB 53. The California law, passed last year, introduced requirements around transparency and protections for whistleblowers at major AI firms. At the time, OpenAI was among those pushing back against the bill.

Now, however, the company is framing state action as a practical path forward in the absence of major federal AI legislation. OpenAI said it supports what it describes as a “reverse federalism” approach, where states advance in broadly compatible ways on core safeguards that could later serve as the basis for a nationwide standard.

The shift suggests that, at least for now, OpenAI sees state governments not simply as regulators to resist, but as potential partners in shaping baseline AI safety rules that could eventually influence policy across the United States.