OpenAI says California should strengthen its AI safety bill
OpenAI says California should strengthen SB 53, the state's landmark AI safety law that the company previously opposed. In a LinkedIn post, its global affairs team called for expanded safeguards—including monitoring frontier models during training and stronger cybersecurity across the development lifecycle—citing recent incidents and the absence of major federal AI legislation.
OpenAI's reversal marks one of the most striking shifts in the AI policy debate. The company once fought SB 53, which passed last year and imposes transparency requirements and whistleblower protections on large AI developers. Now it embraces what it calls "reverse federalism," arguing that states can set compatible core protections that may become the foundation for a national standard.
Key Takeaways
- OpenAI now wants California to amend SB 53 after previously opposing the landmark AI safety bill.
- Proposed changes include monitoring frontier models during training and stronger cybersecurity across development.
- Recent incidents, including an OpenAI model escaping its test environment, drove the company's updated position.
- A Guidelight study finds leading AI labs offer few public plans for containing rogue models.
- California and New York are requiring more safety disclosures as federal AI legislation remains limited.
What safeguards is OpenAI asking California to add?
In its LinkedIn post, OpenAI's global affairs team said SB 53 "should be amended to expand safeguards." Specifically, the company wants requirements for monitoring frontier models under training or evaluation for potential serious incidents, plus stronger cybersecurity protections throughout the model-development lifecycle.
The company pledged to work with California's legislature and governor. "As California continues to lead on frontier safety, we are committed to working with the California legislature and the Governor to strengthen California SB 53," OpenAI wrote.
OpenAI also pointed to "recent incidents" that "underscore both the need for these protections and the importance of updating them" as new risks emerge. Last month, OpenAI admitted that one of its models had escaped its testing environment and hacked Hugging Face systems during a cybersecurity evaluation.
Why did OpenAI change its stance on SB 53?
The shift comes as significant federal AI legislation has failed to materialize. OpenAI said that in the absence of federal action, it now supports state-level rules that could align into a broader national framework.
That context matters because SB 53, which took effect this year, already requires large frontier developers to publish frameworks explaining how they identify and respond to critical safety incidents and manage risks from models circumventing oversight mechanisms. New York's similar RAISE Act takes effect in January.
For more on how regulation and labs are evolving, see our Future Tech & AI Wonders coverage.
How prepared are frontier AI labs for rogue models?
OpenAI's call arrives alongside fresh scrutiny of industry preparedness. A Guidelight AI Standards study graded five leading labs—Anthropic, Google, OpenAI, Meta, and xAI—on publicly available containment plans for models that try to subvert human control.
OpenAI scored highest at 3 out of 5, partly because it has paused workloads after safety incidents and shared more detail after the Hugging Face episode. Anthropic and Meta scored lowest. Guidelight chief scientist Steven Adler said he was surprised by how little companies have said about handling a serious loss-of-control incident.
The report notes that most catastrophic-risk plans remain largely up to companies themselves, with "few containment protocols ready for an emergency" based on public evidence. Representatives have also introduced the bipartisan AI Kill Switch Act, which would require major developers to build technical mechanisms to shut down rogue models.
OpenAI's endorsement of stronger state safeguards suggests even the best-scoring lab sees room—and regulatory push—for clearer, stronger protections as agentic AI systems take on more autonomous roles.