OpenAI is trying to build the rulebook one state capitol at a time, starting with the places most likely to hand it something close to what it wanted from Washington in the first place. Chris Lehane, the company’s chief global affairs officer, has a neat name for the strategy, “reverse federalism.” The label is bureaucratic, but the tactic is blunt: if federal AI law is stuck, shape California, New York, and Illinois until they function like a national model.
That is a smart move if your real goal is not regulation in the abstract, but regulation that arrives with manageable teeth. OpenAI does not need every legislature in America to agree. It needs a few large, politically influential blue states to write similar bills, then let those bills harden into the default setting for everyone else.
The pitch
Lehane’s public argument is easy to understand because it flatters state pride. California, New York, and Illinois are not being asked to wait for Washington to stop squabbling. They are being asked to lead. In Lehane’s telling, these states already care about AI safety, so the real question is whether they will write laws that developers can live with or laws that create unpredictable liability and real financial exposure.
OpenAI’s preferred model is the lighter one. The laws it has backed in California and New York put the emphasis on disclosure and reporting for advanced AI builders. They do not create new broad legal liability for harms caused by frontier systems, and they do not impose the kind of stiff penalties that would make executives genuinely nervous. That is the point. A law can look serious, sound serious, and still leave the industry’s core risk profile mostly intact.
Lehane has been open about the larger ambition. He wants enough major states to copy one another that the country ends up with a single practical standard, whether Congress ever votes on one or not. He also knows the federal route is stalled, and that the tech industry has spent much of the past year trying to convince lawmakers in Washington to keep states from acting on their own because a patchwork would be too messy. OpenAI’s answer is simple. If the patchwork cannot be stopped, buy the patchwork.
The record
California is the clearest proof that this is not just theory. With heavy input from OpenAI lobbyists, Sacramento passed developer rules last year that fit the company’s taste for transparency without much punishment. New York followed with a similar framework. Both states added obligations for large AI companies to disclose and report more, but neither took the next step toward major civil liability or large fines.
Nathan Calvin, who serves as general counsel and vice president for state affairs at Encode AI, says OpenAI has not merely influenced those laws from the outside. He says the company went after an earlier California bill, the one Governor Gavin Newsom vetoed in 2024 after a heavy lobbying push. Calvin also says OpenAI subpoenaed him last year because of his work on California’s AI safety push. That is a company treating state legislation like a live battlefield.
New York offers a second clue. Calvin says OpenAI helped soften that state’s law by pushing Governor Kathy Hochul to keep it close to California’s model. The logic is familiar from other policy fights. If two states are going to regulate, the industry would rather have them regulate in near-identical ways than let one become a source of stronger rules that can spread.
The money around the issue is not subtle either. A pro-AI super PAC network called Leading the Future, partly financed by OpenAI president Greg Brockman, has spent more than $1 million trying to sink the House bid of the New York law’s lead sponsor. Elsewhere, Meta- and Google-backed super PACs have poured $10.7 million into ten California Assembly and Senate races. That is not incidental spending. It is a signal that the AI industry understands state politics as the front line now.
The pressure points
Illinois is where the strategy gets more interesting. The bill moving there does something OpenAI did not seem to expect, it adds mandatory third-party audits for AI developers on top of the transparency and reporting framework California and New York already adopted. Calvin called that a genuine complication, which is a rare moment of plain speaking in a policy fight full of polished euphemisms.
Lehane still expects the Illinois measure to reach Governor JB Pritzker by the end of the state’s May 31 session. If it does, OpenAI will have helped shape three major blue-state templates. That would matter because template laws travel well. Legislators copy what already passed somewhere else. Staffers cut and paste. Governors prefer not to look reckless by rejecting what their peers signed.
Lehane says governors matter more than most people realize, and he is probably right. Legislatures perform the theater, but governors decide whether the script becomes law. Newsom and Hochul are both attuned to a familiar political tension, regulating AI without making their states look hostile to the industry money, jobs, and prestige that come with it. Newsom’s office pushed back hard on the idea that any one stakeholder was steering California’s decisions, and described that claim as false. The denial does not settle the question; it shows the pressure around it.
The larger game
OpenAI says this is about serious risk. The company wants rules strong enough to address the possibility that frontier systems could be used for destructive cyberattacks or other catastrophic harm. That part of the argument is not fake. Frontier model competition has pushed worst-case planning closer to the center of the debate, and regulators do have a legitimate reason to ask what happens when powerful systems are released faster than the state can inspect them.
But the structure of the policy OpenAI prefers tells a more self-interested story. Transparency rules create paperwork. Audits create process. Neither forces the company to accept broad new liability for damage caused by its models. That is a comfortable lane for a firm that wants public reassurance while limiting the legal surface area it exposes.
The privacy analogy is useful here. Tech lobbyists once helped produce nearly identical state privacy laws when federal privacy legislation stalled. Those laws looked like progress, but advocates hated them because they blocked private lawsuits and made industry coordination easier. OpenAI is trying to recreate the same pattern with AI safety, except with a more respectable moral wrapper.
OpenAI does not need Republican unanimity for the plan to work, but it does need Republican targets. Lehane has said the company cannot pull this off without eventually dealing with GOP lawmakers and governors. That matters because red states are not blank slates. Some conservative lawmakers will reflexively dislike new regulation, while others are already focused on child safety, fraud, and the way AI systems are being used in schools and homes. OpenAI has already endorsed Senator Marsha Blackburn’s Kids Online Safety Act, a signal that it understands where the opening may be. In those states, the debate may land less on frontier model risk and more on child protection.
There is another variable in the background, the White House. Lehane says red-state leaders are watching to see what the Trump administration does next, especially after recent cybersecurity concerns tied to Anthropic’s Mythos AI model. If the administration issues an executive order tightening controls on advanced AI development or deployment, Republican-led states may suddenly have cover to move.
Congress still sits at the center of the whole operation, even if it is not behaving like it. Lehane is betting that a cluster of blue-state laws could become the argument for a federal standard later, especially if Democrats regain power in Washington. That is the real shape of the plan. Use the states to create momentum, then use the momentum to box in Congress.
If it plays in Peoria, Lehane says, it should play in Washington. That is one way to put it. Another is simpler. OpenAI is trying to write the national law without admitting it is writing the national law.
