OpenAI wants California to strengthen its new AI safety law
OpenAI wants California to strengthen SB 53, the AI safety law Governor Gavin Newsom signed in September 2025. Less than a year after helping pass the Transparency in Frontier Artificial Intelligence Act, the company is urging new monitoring requirements and stronger cybersecurity rules for frontier AI models during training and evaluation. The push signals that even a law OpenAI once backed may not go far enough as models grow more capable.
The request came in a LinkedIn post from OpenAI's Global Affairs team. OpenAI says California should require frontier AI models to be monitored during training or evaluation for signs they could bypass a third party's security controls or obtain confidential information they should not have access to. The company is also calling for beefed-up cybersecurity across the entire model development process, aimed at preventing AI systems from circumventing their internal safeguards.
Key Takeaways
- OpenAI wants California to amend SB 53 with stronger monitoring and cybersecurity requirements for frontier AI models.
- SB 53, signed in September 2025, already requires safety frameworks, incident reporting, whistleblower protections, and civil penalties.
- OpenAI frames its push as 'reverse federalism,' urging states to lead while Congress has not passed comprehensive AI legislation.
- The company insists the ask is forward-looking, not tied to a single incident, though it keeps OpenAI in the room to shape future standards.
- California created CalCompute and mandated annual law reviews, reflecting how quickly frontier AI capabilities evolve.
Why does OpenAI want California to strengthen SB 53 now?
The timing matters. Governor Gavin Newsom signed SB 53 into law on September 29, 2025, and OpenAI was among the voices that helped get it across the finish line. Now, with frontier models advancing faster than many policymakers expected, the company argues the statute needs new teeth.
Specifically, OpenAI wants requirements that frontier AI models be watched during training or evaluation for behavior that could evade external security controls or pull confidential data from systems they were never meant to reach. That goes beyond the transparency and reporting framework SB 53 already established.
OpenAI was careful to frame the request as proactive rather than reactive. The company said the goal is not 'to write rules for one particular incident' but to build a framework that catches problems earlier and ensures lessons are shared industry-wide. Still, the move keeps OpenAI at the table as California decides what stronger enforcement should look like.
What did California's AI safety law require when it passed?
When SB 53 landed on Newsom's desk, it was already a significant piece of legislation. According to the Governor's office, the Transparency in Frontier Artificial Intelligence Act requires large frontier AI developers to publish safety frameworks on their websites. It also establishes a formal channel for reporting critical safety incidents to California's Office of Emergency Services.
The law protects whistleblowers who raise concerns about AI safety and grants the state Attorney General authority to levy civil penalties on companies that do not comply. It created CalCompute, a public computing consortium meant to boost safety and equity research. California also committed to revisiting the law annually as the technology beneath it shifts.
Newsom pitched SB 53 as proof that California could protect the public without choking off the AI industry that has become a major part of the state's economy. At the time, he called it a stopgap for Washington's inability to pass anything resembling real federal AI policy. For a deeper look at how tech regulation has shifted over the years, explore our Nostalgia: Then & Now coverage.
How does SB 53 compare to federal AI policy today?
Then and now, the federal picture has barely moved. Congress still has not passed comprehensive AI legislation, leaving states to fill the vacuum. California's SB 53 was designed as a model other states could follow, and OpenAI's broader strategy even has a name for this moment: 'reverse federalism.'
The idea is straightforward. While Washington sits on its hands, states can move first on compatible safety standards that could eventually become national policy. OpenAI is betting that California, home to much of the AI industry, is the best place to start tightening rules before a crisis forces a more chaotic response.
That dynamic mirrors an older pattern in American tech regulation. States often experiment first — think privacy, emissions, or consumer protection — and federal law catches up years later. SB 53 fits that playbook, but the speed of AI development means the gap between state experiment and outdated rulebook is shrinking fast.
What is OpenAI's 'reverse federalism' strategy?
OpenAI's Global Affairs team is not just asking for amendments in Sacramento. The company is making a broader argument about how AI governance should work in a country where federal action has stalled. By pushing California to strengthen a law it already supported, OpenAI signals that voluntary commitments and post-deployment transparency may not be enough.
The requested changes focus on two areas: continuous monitoring during model training and evaluation, and stronger cybersecurity throughout development. Both aim to stop AI systems from bypassing safeguards before a model ever reaches the public. That is a shift from SB 53's emphasis on published frameworks and incident reporting after deployment.
Whether California lawmakers take up OpenAI's proposals remains to be seen. The annual review built into SB 53 gives the state a natural moment to revisit its requirements. For now, the headline is clear: a company that helped pass California's flagship AI safety law now wants that same law to bite harder — and it wants the rest of the country to watch what happens next.