Federal Preemption of State AI Laws: Is the NDAA Next?

News Published: 10 min read Pravesh Garcia
Federal Preemption of State AI Laws: Is the NDAA Next?

If a chatbot gives your teenager dangerous advice, whose rules did it break? Right now, that depends on which state you live in. That’s the heart of the fight over federal preemption of state AI laws. Washington wants one national rulebook. The states keep writing their own.

Every few months, someone tries to settle the matter with a shortcut. They tuck a ban on state rules into a bill that has nothing to do with chatbots. The newest version of the story says preemption is heading back into the National Defense Authorization Act, the annual defense bill.

Maybe. As of November 11, 2026, I couldn’t find a single confirmed report of preemption language in the FY2027 NDAA. I couldn’t find scheduled talks to add it, either. What the record does show is a pattern: serious attempts to sideline the states, and none of them has delivered what its backers wanted.

So here’s where things actually stand, what’s confirmed, and what’s still a forecast.

What federal preemption of state AI laws actually means

Preemption sounds technical. The idea is simple. When Congress passes a valid federal law, it can override state laws that conflict with it. Applied here, federal preemption of state AI laws would mean a Colorado or California rule stops binding companies where it clashes with the national one.

The catch sits in the details. The first big attempt wasn’t a national rulebook at all. It was a pause: years during which states couldn’t enforce their own AI rules.

A pause isn’t a standard. That distinction turned a dry legal question into a political brawl.

In practice, three routes decide who regulates AI in the US:

  • Congress can pass a law that preempts the states outright.
  • The White House can pressure states with lawsuits and money, but it can’t erase a state law by decree.
  • State legislatures can keep passing laws until one of the other two stops them.

All three are live. So far, only the third has produced new AI laws.

Three attempts, three stumbles

The 99-1 Senate vote (July 2025)

The first serious attempt rode on the 2025 budget bill. It began as a 10-year pause on state AI laws, and negotiators cut it to five. Then the Senate voted 99-1 to strip it out entirely.

Ninety-nine to one. You almost never see that margin on anything contested.

The opposition didn’t split along party lines. Sen. Josh Hawley called it “a terrible provision.” Republican governors Ron DeSantis and Sarah Huckabee Sanders opposed preemption too.

The FY2026 NDAA gambit (late 2025)

Backers tried again that fall, this time with the defense bill. The logic was easy to follow. Congress treats the NDAA as must-pass, so riders that can’t survive alone sometimes hitch a ride.

The President leaned in. On November 18, 2025, Trump posted on Truth Social, as IAPP reported: “We MUST have one Federal Standard instead of a patchwork of 50 State Regulatory Regimes.”

It still failed. More than 290 state lawmakers wrote to Congress opposing a ban on state AI laws. That count comes from the advocacy group Americans for Responsible Innovation. Faith leaders, unions and whistleblowers sent letters of their own.

In early December, House Majority Leader Steve Scalise confirmed the language wouldn’t go into the FY2026 NDAA. The defense bill, he said, “wasn’t the best place.”

Brad Carson, ARI’s president, read the result as a verdict. “Now Congress has gotten the message: Americans want safeguards that protect kids, workers, and families,” he said.

But listen to the rest of what Scalise said. The language could show up “in other places.” Hold onto that phrase. It’s the reason people keep checking the defense bill.

Executive Order 14365 (December 11, 2025)

When Congress balked, the White House moved on its own. On December 11, 2025, President Trump signed Executive Order 14365, “Ensuring a National Policy Framework for Artificial Intelligence.” It does four main things:

  1. Sets up a DOJ AI Litigation Task Force. The Attorney General had 30 days to create it. Its “sole responsibility,” in the order’s words, is to challenge state AI laws that conflict with the administration’s policy. Grounds include interstate commerce and preemption.
  2. Orders a Commerce Department review of “onerous” state AI laws, due within 90 days.
  3. Puts money on the table. States with onerous laws could lose remaining non-deployment funds under BEAD, a federal broadband program. That applies only as far as federal law allows.
  4. Calls for draft legislation that would preempt conflicting state laws. It spares child safety rules and AI compute and data center infrastructure (except generally applicable permitting). It also spares how states buy and use AI inside their own governments.

Danielle Davis of the Joint Center reacted while the order was still a draft. She called a dedicated unit whose sole purpose is challenging state AI protections “extraordinary.” It’s hard to argue with the adjective.

What the executive order and DOJ task force can’t do

Here’s what the headlines tend to skip. The order doesn’t achieve federal preemption of state AI laws on its own.

Lawyers at Crowell & Moring put it bluntly: the order “does not have the force of law nor is it self-executing.” It never defines “onerous,” either. That leaves the most important word in the document open to argument.

Two legal walls stand in the way, too. Courts tend to resist finding preemption in agency action rather than a statute from Congress. And the Tenth Amendment’s anti-commandeering doctrine limits how hard Washington can push states to do its bidding.

So the order works through pressure, not command. The task force can sue. Commerce can name names. The funding threat can make a state budget office sweat. But a state law stays on the books until a court strikes it down or Congress overrides it.

Has the pressure worked? The evidence is thin. As of early July 2026, Model Diplomat reported that no major DOJ suit had produced a clear win. That’s one outlet’s read. I couldn’t confirm what the task force has done since, or whether Commerce ever published its list of onerous laws. If someone tells you the administration has already won in court, ask to see the ruling.

The state patchwork is smaller than the rhetoric

“A patchwork of 50 states” makes a good soundbite. The numbers tell a messier story.

By July 1, 2026, 29 states had enacted 109 AI laws, plus 28 laws on data centers. That’s from a tally by the NYU Center on Technology Policy, reported via Tech Policy Press and Model Diplomat. At the same point in 2025, the count stood at 121 laws across 39 states.

Read that again. Fewer laws, and fewer states passing them.

Two caveats before anyone builds an argument on it. It’s a mid-year snapshot, and legislatures don’t all run on the same calendar. I’ve also only seen it through one outlet’s reporting, not NYU’s original dataset.

One reading says federal pressure cooled some statehouses. Another says 2025 was a frantic first wave and 2026 a natural slowdown. The count alone can’t tell you which, and I’d distrust anyone who claims it can.

The bill pipeline is another matter. The draft version of the executive order cited more than 1,000 AI bills introduced in state legislatures. Most bills die. The survivors include the laws this fight keeps circling back to:

  • The Colorado AI Act
  • California’s Transparency in Frontier AI Act (SB 53), focused on frontier AI developers
  • New York City’s bias audits and Illinois’s notice and bias rules

Model Diplomat’s reporting also flags where much of 2026’s state activity landed: chatbots, kids and data centers.

Is preemption really back in the NDAA?

Here’s the honest ledger, as of November 11, 2026.

Confirmed:

  • The Senate voted 99-1 in July 2025 to strip a pause on state AI laws from the budget bill.
  • Preemption language stayed out of the FY2026 NDAA in December 2025.
  • Scalise said the language could appear “in other places.”
  • President Trump signed EO 14365 on December 11, 2025. It set up a litigation task force, a Commerce review and funding conditions.

Not confirmed:

  • Any preemption text in the FY2027 NDAA.
  • Any scheduled late-November negotiation to add it.
  • Results from DOJ task force lawsuits after early July 2026.
  • Whether Commerce published its “onerous” list.
  • Reported White House legislative recommendations from March 2026. Coverage mentions them, but I couldn’t verify them from the source documents.

So why does the NDAA rumor keep coming back? Because the incentives haven’t changed. The defense bill is one of the few vehicles Congress reliably passes, and Scalise left the door open. Some policy watchers have floated the FY2027 bill as a possible home for another try. That’s a prediction, not a report.

The law firm Fisher Phillips made a sharper forecast after the December drop. It expected any future bill to be narrower, built around national competitiveness or critical infrastructure. It also expected carve-outs for discrimination and safety.

I think that’s the likelier shape. A blanket ban lost 99-1. A narrow bill with real exceptions might not.

Why this fight reaches past the lawyers

It’s tempting to file all this under turf wars. Don’t.

Look at what state AI rules tend to cover: kids and chatbots, bias in automated systems, transparency from frontier model developers. Those protections are on the line whenever someone pushes federal preemption of state AI laws with no federal floor behind it.

Even the White House framework seems to know where the lines are. It says a national standard shouldn’t override state child-safety laws or data center infrastructure rules. States would also keep control over their own use of AI. That list tells you which fights the administration expects to lose.

The data center carve-out deserves a closer look than it usually gets. Those buildings sit in real towns and draw on real power grids, which is part of why we’ve looked at why AI data centers are turning to nuclear power.

The patchwork complaint isn’t empty, though. If you ship one product nationwide, juggling dozens of state rules costs real money. Big firms absorb that cost more easily than small ones. A single standard could help smaller players, if it actually protects people.

Other governments have answered this question differently. Europe went with one binding law, as we covered in our guide to how AI is regulated under the EU AI Act. The US hasn’t settled even the first question: who gets to write the rules. And that’s before you reach the harder ones we explored in AGI regulation and superintelligence planning.

What to watch next

  • FY2027 NDAA text. Committee drafts and the final conference text are where a preemption rider would surface first.
  • Court rulings in task force cases. One clear decision would tell us more than a year of speeches.
  • The Commerce list. If it appears, “onerous” finally gets a working definition, and states get a target on their backs.
  • BEAD funding decisions. The first state that actually loses money will show whether the lever is real.
  • The 2027 state sessions. Does the count of new state AI laws keep falling, or was 2026 a breather?

Here’s where I land. The real question was never Washington versus the states. It’s whether anyone writes rules you can count on.

A national standard with genuine protections would be a fair trade. A ban on state rules with nothing behind it would be a vacuum with a flag planted in it. Until Congress acts, statehouses are where AI laws actually get passed.

If preemption language turns up in a defense bill this winter, skip the press release and ask one question. What replaces the rules it erases?

Frequently Asked Questions

It means a federal law overriding state AI laws that conflict with it. Only Congress can do that by statute, and so far its attempts have failed: the Senate voted 99-1 in July 2025 to strip a pause on state AI laws from the budget bill, and preemption language was dropped from the FY2026 NDAA in December 2025.
Not directly. Executive Order 14365, signed December 11, 2025, created a DOJ AI Litigation Task Force, ordered a Commerce Department review of 'onerous' state AI laws, and tied some federal broadband (BEAD) funds to state policy. Lawyers at Crowell & Moring note the order does not have the force of law and is not self-executing, so a state law stays in place unless a court strikes it down or Congress overrides it.
Opposition was broad. Hundreds of state lawmakers wrote to Congress against a ban on state AI laws (Americans for Responsible Innovation counted more than 290), alongside faith leaders, unions and whistleblowers. Republicans including Sen. Josh Hawley and Govs. Ron DeSantis and Sarah Huckabee Sanders opposed it too. House Majority Leader Steve Scalise said in December 2025 that the NDAA 'wasn't the best place' for it, while saying the language could appear 'in other places.'
As of July 1, 2026, 29 states had enacted 109 AI laws, plus 28 data center laws, according to an NYU Center on Technology Policy tally reported via Tech Policy Press and Model Diplomat. That is down from 121 laws across 39 states at the same point in 2025. It is a mid-year count, so the full-year total may differ.
It is a unit Executive Order 14365 told the Attorney General to create within 30 days. Its sole responsibility, per the order, is to challenge state AI laws that conflict with the administration's AI policy, on grounds including interstate commerce and preemption. As of early July 2026, Model Diplomat reported that no major DOJ suit had produced a clear win.
Not as far as anyone has confirmed. As of November 11, 2026, we found no verified reporting of preemption language in the FY2027 NDAA or of scheduled negotiations to add it. Some observers expect another attempt, partly because Steve Scalise said in 2025 the language could appear 'in other places,' but that remains a forecast.