Six months ago President Trump told the states to stand down on artificial intelligence, warning that a patchwork of local rules would hobble the industry in its race with China. The states did not listen. More AI bills have been introduced this year than last, sponsored by Republicans as well as Democrats, and a string of them have become law while Congress stays stalled. As one Illinois sponsor said after brushing aside the president's threat, "I don't know if you've met Illinois, but we're pretty independent."
The new wave is narrower and more practical than the sweeping anti-bias bills governors vetoed in earlier rounds. Lawmakers have zeroed in on the places where ordinary people meet AI without realising it. Colorado now requires companies to tell you when an AI system is helping decide your job application, loan or housing. Connecticut has restricted companion chatbots that form ongoing relationships with minors. A mix of red and blue states, Idaho, Iowa, Nebraska and Oregon among them, have passed rules forcing disclosure when you are talking to a machine rather than a person. California is weighing a "No Robo Bosses Act" to stop employers firing workers on an algorithm's say-so.
Illinois went further, onto the frontier itself. Senate Bill 315, on Governor Pritzker's desk, borrows from California and New York laws requiring developers of the largest models to maintain protocols against catastrophic misuse, then adds an independent auditor to check the companies actually follow their own rules. It passed with near-unanimous support, a sign that filling the federal vacuum has become a bipartisan instinct rather than a partisan one.
Washington's answer is to try to take the pen back. Axios reports the White House is negotiating to preempt some state AI laws in exchange for backing congressional priorities such as the Kids Online Safety Act, with Senator Marsha Blackburn leading the talks. A separate bipartisan House draft, the Great American AI Act from Representatives Obernolte and Trahan, would impose a three-year freeze on state rules governing model development while creating federal safety-reporting duties for labs earning more than 500 million dollars. Both efforts have already drawn fire from every direction: conservative Chip Roy calls preemption a "non-starter," while consumer advocates warn the bill "replaces a state floor with a federal ceiling."
The stakes turned concrete on Friday. A coalition of state attorneys general opened an investigation into OpenAI, with New York's Letitia James subpoenaing documents on advertising, minors, health data and model "sycophancy." OpenAI said it would "engage constructively." The timing was almost too neat: at the exact moment Washington debates how much authority to strip from the states, the states are using that authority to interrogate the industry's most prominent company. Whether their subpoenas survive depends entirely on where a future bill draws the line between ordinary consumer protection and AI-specific rules.
That line is the whole fight, and it does not exist yet. There is no bill text, no agreed scope, no signed deal. Until there is, the practical reality is the inverse of the one Trump described last winter: not a single federal standard holding the field, but fifty legislatures writing their own, faster than Congress can decide whether to stop them. For AI companies hoping for "one national standard," the patchwork they feared is the thing arriving first.