Everyone Has Ideas About AI Regulation in the New Congress
This post is part of my Medium blog.
Last week I wrote that I don't expect a Democratic Congress to pass much AI law in the next two years. I still believe that. This week, though, everyone headed to Washington seems to have an idea about how to regulate AI, and one of the bills already sitting in Congress says something about the states that almost nobody is talking about.
Here's who to watch
The Hill profiled five incoming Democrats the AI safety crowd is excited about. Scott Wiener is taking Nancy Pelosi's seat in San Francisco. He wrote SB 1047, the California frontier-model bill Newsom vetoed, and then SB 53, the transparency law Newsom signed. Abdul El-Sayed won the Michigan Senate race on a platform that would require every frontier AI lab to be chartered as a public benefit corporation, and he wants an FDA-style agency to test models before release.
A public benefit corporation is still a for-profit company with shareholders. The difference is that its charter names a public mission, and the board is legally required to weigh that mission alongside profit instead of answering only to returns. It gives directors cover to put the mission first more than it forces them to. And some of the biggest labs are already there. OpenAI's for-profit arm became OpenAI Group PBC in October 2025, still controlled by its nonprofit foundation, and Anthropic has been a PBC from the start. What El-Sayed adds is the requirement: every frontier lab, by law, with no option to change its mind later.
Micah Lasher, headed to New York's 12th District, co-sponsored New York's RAISE Act. Manny Rutinel from Colorado and Sam Forstag from Montana round out the list. Forstag put the case most bluntly in a September video: Congress is "moving too damn slow."
Then there's Illinois Gov. JB Pritzker, who is clearly getting ready to run for president. The day after he won a third term, he released a 200-point agenda for the new Congress, a 75-page document the Chicago Sun-Times said "could easily be mistaken for a presidential platform." He hasn't announced anything, but governors don't usually write legislative programs for a Congress they don't sit in. The AI section asks for federal requirements on AI risk assessment, testing, and monitoring. According to WTTW and Capitol News Illinois, he also said any federal law should build on what Illinois and other states already did, not override it. Remember that part.
None of the new members take office until January. Until then, the Republican majority runs the calendar, and it has a lot on it.
The unfinished bills
Congress came back Monday for the lame-duck session, and Nextgov's rundown of what's waiting is mostly deadlines. On December 11 the 2015 law that protects companies who share cyber threat information with the government expires again, and government funding and the stalled defense bill need the same few weeks.
The AI safety bills are further back in line. There's the AI Kill Switch Act from Ted Lieu and Nathaniel Moran, which would make big developers keep the ability to shut their systems down and let DHS order emergency action after an incident. Sanders and Casar want to ban superintelligence outright. And there's the FRONTIER Act, from Republican Jay Obernolte and Democrat Lori Trahan. That's the one to read.
Read the bill
The FRONTIER Act is H.R. 9925, introduced July 23 with sponsors from both parties, and it has real teeth. Any company training a model above ten to the twenty-sixth operations, with serious revenue, would have to publish a safety framework, hire an independent auditor every year who can't be paid based on the result, publish a transparency report before each release, and report a critical safety incident within 72 hours. An incident that threatens lives goes to law enforcement within 24. Penalties run up to a million dollars a violation, per day, and the Commerce Secretary could order a dangerous model suspended.
If you've followed the state laws, this is SB 53 and the RAISE Act, scaled up and made federal.
Then you get to Section 9. It says no state "may adopt or enforce any law, regulation, order, or other requirement that imposes new substantive obligations on artificial intelligence developers" in three areas: frontier risk transparency, third-party auditing, and incident reporting. Those are exactly what California, New York, and Illinois spent the last two years writing into law. States could still regulate how AI gets used, protect kids, and set their own purchasing rules, but the frontier labs themselves would answer to Washington alone. Lawyers will argue about what "new" means.
On its face, it's a trade: a strong federal rule in exchange for the state rules.
Mira read the amendments
In Essential, there's a scene where Mira is two doors down from Rob, doing something rare in 2033: reading the amendments, pencil in hand, while everyone else works from their agent's summary. I wrote that as a small character note. Section 9 is the reason it matters.
A bill that isn't passed by January 3 dies with this Congress, so the FRONTIER Act will almost certainly have to be reintroduced, with the five people The Hill profiled in the room. Two of them wrote or co-sponsored the state laws Section 9 would push aside. Pritzker, who signed the Illinois version, has already said he doesn't want them overridden.
So the questions for January are specific. Does the new bench keep the audits, the 72-hour reporting, and the emergency stop? Does it keep Section 9? And will a President who spent two years trying to stop the states sign a bill that lets them keep their rules? Watch the December 11 deadline first, then watch who reintroduces this bill and what the last section says.