The Blocker Is Gone. The States Were Already Writing the Rules.
This post is part of my Medium blog.
Two days ago I wrote about what the midterms changed. One thing I left out, because it deserved its own post: the two-year fight over who gets to regulate AI, and why it just ended in a way almost nobody is talking about.
Here's the pattern of the last two years, stated plainly. The states were ready and willing to regulate AI. The Republican-led Congress kept trying to stop them. That was the actual fight — not whether AI should be regulated, but whether the states would be allowed to. And Congress lost. Twice, in public, with the votes counted.
The first attempt was a ten-year moratorium on state AI laws, tucked into a budget bill in the summer of 2025. The Senate stripped it out 99 to 1. That is not a close vote. That is the entire Senate, both parties, telling the drafters that states do not lose their police power because a provision was written into a budget reconciliation bill. The second attempt came a few weeks ago, when word got out that the same moratorium had been attached to the defense authorization bill. It was quietly removed before the text went public. The House Majority Leader admitted the defense bill "wasn't the best place" for it, and Senator Josh Hawley — not exactly a states'-rights moderate — posted that it was a terrible provision and should stay out. When your own party's senators celebrate the removal of your own bill, the fight is finished.
The White House has one more arrow, and it already fired it. In December 2025 the President signed the "one rulebook" executive order: a Justice Department task force to sue states over their AI laws, threats to withhold federal funding, and a demand that Congress pass a national framework that preempts the states. A year later, no state AI law has been struck down, the funding threats are tangled in litigation, and the legislative recommendation has gone nowhere. An executive order can't do what Congress declined to do. Preemption is a legislative power. The courts will say so, eventually, and probably at length.
Now the election has removed the blocker entirely. But be honest about what replaces it, because I've seen the premature takes already. A Democratic Congress is not going to pass aggressive AI legislation on its own. Big bills need sixty votes in the Senate or a coalition of Republicans willing to cross over, and after the week this party just had, I can't rule anything out — but I wouldn't bet on it. And every bill that somehow clears both chambers still lands on the desk of a President who spent 2025 and 2026 fighting to stop states from regulating at all. He holds the veto pen. I do not expect the legislative branch to make much progress on AI law in the next two years.
Here's the part people are missing. That doesn't mean nothing happens. It means what happens is already happening, somewhere else.
The states never waited for permission. The rules below are not proposals. They are law, most of them already in force.
California went first and heaviest. SB 53, the Transparency in Frontier Artificial Intelligence Act, took effect January 1st. It covers the biggest developers — the ones training models above ten-to-the-twenty-sixth operations — and requires them to publish safety frameworks, report serious safety incidents to the state within fifteen days, and protect whistleblowers who come forward. Violations can cost a million dollars each for the largest companies. Two companion laws require generative AI developers to disclose what went into their training data, and require watermarks or clear disclosure on AI-generated content. New York followed with the RAISE Act, signed in December and amended into its final form in March. It tracks the California model — transparency obligations on frontier developers — and hands rulemaking to the Department of Financial Services. It takes effect January 1, 2027.
Illinois attacked a different layer. HB 3773 makes it a civil rights violation to use AI in hiring, promotion, discipline, or firing in a way that discriminates, or without telling employees it's being used. In force since January. This summer the governor signed SB 315, which puts audit and disclosure obligations on the largest frontier developers starting in 2027.
And the red states are in this too, which is the part that breaks the simple story. Utah passed the country's first generative AI consumer protection law back in 2024 — businesses have to disclose when you're talking to a bot, proactively in high-risk situations like medical, legal, or financial advice. Texas TRAIGA has been in force since January: it prohibits using AI to deliberately manipulate people or unlawfully discriminate, and gives companies a safe harbor if they follow the NIST risk management framework. Colorado passed the most ambitious law of all in 2024, watched the industry melt down over it, delayed it twice, and finally repealed and replaced it with a narrower automated-decision law taking effect in 2027. That's states doing what states do — legislating, stumbling, revising. Nineteen new state AI laws were signed in the last two weeks of March 2026 alone. There are thousands of AI bills in play across the fifty statehouses.
So the map for the next two years is this: no federal statute, an executive branch still trying to sue the states into stopping, and dozens of state governments writing rules at their own speed, in their own directions. For anyone shipping AI products nationally, that's the hard scenario. Fifty compliance lanes instead of one. The companies spent two years lobbying Congress for a single rulebook because they said a patchwork would kill innovation. Congress declined to save them, and now the patchwork is what they get.
In the books, this is roughly how it goes. Redundant and Contingent run on the premise that governments respond to AI late, unevenly, and locally — that regulation arrives as a scramble of jurisdictions rather than one clean law. I didn't write that because I'm clever. I wrote it because it's the American pattern. We regulated cars, food, drugs, and aviation this way: the states move, the feds follow a decade later, and the companies operate in the gap.
Watch these dates instead of Washington. January 1, 2027: New York's RAISE Act, Illinois SB 315, and Colorado's replacement law all switch on in the same month, plus California's automated-decision rules. Before that, watch the courtroom where the Justice Department's task force meets its first state AI law. The action moved to the statehouses, and it's not moving back.