ANALYSIS The two leading frontier AI labs are no longer debating whether states should regulate AI — they are competing to shape which regulatory model wins. OpenAI now backs strengthening California's SB 53 while opposing Massachusetts' stricter proposal; Anthropic supports the Massachusetts framework while endorsing transparency laws elsewhere. The divergence reveals that the real contest in US AI safety regulation is not industry versus government but lab versus lab, each maneuvering to lock in the oversight regime that best fits its own competitive position.
Why it matters
With no significant federal AI legislation in place, state capitals have become the primary arena for AI governance. OpenAI explicitly endorsed "reverse federalism," arguing that "states can move in a compatible direction around core protections that can ultimately become the foundation for a national standard"1. ◆ Whichever state model gains the most traction could effectively set the template for national regulation — making each lab's state-by-state positioning a high-stakes bet on the future regulatory landscape.
The big picture
OpenAI is calling for California to amend SB 53 — the landmark AI safety law passed last year that imposes transparency requirements and whistleblower protections on large AI companies — to "expand safeguards". Specifically, the company wants amendments "requiring monitoring of frontier models under training or evaluation for potential serious incidents" and "strengthening cybersecurity protections throughout the model-development lifecycle". The company referenced "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.
OpenAI's endorsement is striking because it previously opposed SB 533. Yet in Massachusetts, where a proposal described as "the nation's most stringent state-level safeguards" for AI is moving through the legislature, the two labs have traded places4. The Massachusetts bill would allow outside organizations to evaluate a model's potential dangers every 120 days, based on their own guidelines. OpenAI's head of U.S. state policy and partnerships, Donnie Fowler, indicated that OpenAI prefers states model their standards on the Illinois law, which requires a third-party audit once a year to ensure labs are following their own safety guidelines.
Anthropic, by contrast, backs the Massachusetts approach. Anthropic's head of U.S. state and local government relations, Cesar Fernandez, said the more intensive third-party evaluations are needed because "we ultimately don't think the industry should grade its own homework". Anthropic said it has supported newly passed laws in several states that require greater transparency for AI developers.
ANALYSIS The fault line is not about whether regulation should exist but about who controls the evaluation criteria. California's SB 53 framework, even with OpenAI's proposed amendments, keeps monitoring tied to the model-development lifecycle — an area where the developer retains significant control over what is disclosed and when. The Massachusetts model shifts authority outward: external organizations set their own evaluation guidelines and publish findings publicly, though the state would not be able to halt AI development.
OpenAI's reversal on SB 53 becomes more legible in this context. By championing a California framework it can help shape — and proposing specific amendments — OpenAI positions the SB 53 model as the credible "responsible" standard, potentially preempting the Massachusetts-style approach from spreading. The company's lobbying team has been working directly with bill sponsors to propose specific amendments.
Anthropic's stance carries its own strategic logic. Backing the strictest available framework reinforces a brand built on safety-first positioning and raises the compliance burden for competitors who may be less prepared for frequent external audits. The Massachusetts proposal is included in a larger economic development bill that was passed by the Massachusetts Senate last month but still requires approval by the House and the governor.
The Massachusetts proposal would go further than the AI safety laws adopted by California, New York, and Illinois. ANALYSIS Each new state law that passes creates a precedent other legislatures can adopt or adapt, meaning the OpenAI-Anthropic split is not merely a policy disagreement — it is a race to establish the dominant regulatory template before federal action materializes.
What's next
The Massachusetts bill still requires approval by the House and the governor. California's legislature continues to debate SB 53 amendments, with OpenAI saying it is "committed to working with the California legislature and the Governor to strengthen" the law. ◆ The outcome of these two legislative processes will signal which regulatory model — developer-guided monitoring or independent external evaluation — gains momentum as the de facto state-level standard. The labs' public positioning makes clear that neither intends to leave that choice to legislators alone.