A Democratic Congress Can't Pass an AI Law on Its Own. Here Is What It Can Change.

Democrats won control of both the House and the Senate in the 3 November 2026 midterms, according to CBS News projections; final seat counts are not in. That majority will not write America's AI law on its own: any bill still needs President Trump's signature, and the rule he has asked Congress for, one national standard that overrides state AI laws, is the one many Democrats have spent a year fighting. What changes is narrower, and it starts on 3 January. The plan to preempt state AI laws through Congress, which already failed twice under a Republican majority, loses the votes it would need, and Democrats take the committee gavels that run hearings, subpoenas and budgets.

The projection is CBS News's, on its live page stamped 4 November 2026 at 2:38 AM EST: "CBS News projects that Democrats will win control of the House and Senate in the 2026 midterm elections." The piece rests on the executive orders, two bills, the Congressional Record, the White House's recommendations and the states' own announcements, read in the original; members' words, and two news reports, come from the outlets named beside each. One finding runs through all of it. The broad plan to override state AI law loses its votes. A narrow version does not, and it is already written. H.R. 9925, the FRONTIER Act, introduced on 23 July 2026 by a Republican with three Democratic cosponsors, would bar states from imposing new obligations on AI developers in three areas: frontier risk transparency, third-party auditing and incident reporting. Nothing in its text makes that bar expire.

Can a Democratic Congress repeal Trump's AI executive order?

No, not without the President's signature or a veto override. Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence", signed 11 December 2025 and published at 90 FR 58499, stays in force until the President revokes it, a court stops it, or a law overrides it. A bill the President returns with objections becomes law only "if approved by two thirds of that House", in the Constitution's words, in each chamber in turn.

The repeal bill already exists. H.R. 8031, the GUARDRAILS Act, introduced on 20 March 2026 by Rep. Beyer with Democratic cosponsors including Rep. Lieu, Rep. Khanna and Rep. Tlaib, has one operative sentence: "The Executive order entitled ``Ensuring a National Policy Framework for Artificial Intelligence'', issued on December 11, 2025, shall have no force or effect, and no Federal funds may be used to implement, administer, enforce, or carry out that Executive order." A Democratic House can pass that sentence. In the Senate it first meets cloture, the vote to end debate, which the Congressional Research Service puts at "three-fifths of the Senators duly chosen and sworn," "or 60 votes if there are no vacancies in the Senate's membership." Then it goes to the President whose order it voids.

So the order keeps running after 3 January. Here is what it runs, in its own words.

When we read those directives against the public record in September, the only one we could show had produced binding action was the litigation task force; that reading is in Who Limits AI in America, and nothing in this election changes it. The order's own section 9(b) says it "shall be implemented consistent with applicable law and subject to the availability of appropriations." My reading: that clause, not repeal, is where a Democratic Congress touches the order. But a spending law is still a law, and it needs the same signature.

Can Congress still pass a law that preempts state AI laws?

Not the broad version. Broad preemption has already failed twice, in two different bills, with Republicans holding both chambers, and on my reading a Democratic majority leaves it no route at all. The record:

The section the Senate struck is the one usually described as a moratorium on state AI law tied to broadband money; the Digest itself describes it no further than the two lines above. The White House recommendations are explicit about scope: "States should not be permitted to regulate AI development, because it is an inherently interstate phenomenon with key foreign policy and national security implications." That is a plan to take a whole field from the states. Nothing of that reach has passed a Republican Congress; the two earlier attempts above both failed. I see no reading of the new majority under which it passes this one.

What does the FRONTIER Act do to state AI laws?

It would forbid states to impose new duties on AI developers in three named areas, and as introduced it does so permanently. The bill's header reads: "Mr. Obernolte (for himself, Mrs. Trahan, Mrs. Houchin, Mr. Peters, Mr. Scott Franklin of Florida, and Mr. Subramanyam) introduced the following bill". Trahan, Peters and Subramanyam are Democrats. It was referred to the Committee on Energy and Commerce and the Committee on Science, Space, and Technology.

Section 9 works in three steps:

• The bar, 9(b): "no State or political subdivision of a State may adopt or enforce any law, regulation, order, or other requirement that imposes new substantive obligations on artificial intelligence developers with respect to any Covered Subject Area."
• The three areas, 9(a): "Frontier ai risk transparency", "Frontier ai third-party auditing and independent verification" and "Frontier ai incident reporting".
• What states keep, 9(c): generally applicable law; regulation of use and deployment by deployers and users, "provided that no substantive obligations are imposed on developers with respect to model development, training, evaluation, or release"; protection of minors; and state procurement.

Nothing in the introduced text sets an end date. A search of the Government Publishing Office text for "sunset", "expire", "cease to", "2029" and "2030" returns nothing. That is a change from the June discussion draft, about which Rep. Trahan told Tech Policy Press: "Preemption expires in December 2029 assuming passage later this year."

Now set Section 9 beside the two state frontier laws, as the governors' own announcements describe them.

My reading, and it is a reading of the text, not a prediction of any court: Section 9 covers the same subjects as both state laws, and it bars states not only from adopting such rules but from enforcing them. Whether that reaches laws already on the books turns on what counts as a "new" obligation, and that is the word a court would have to settle.

Two features of the text are worth reading closely. First, who is covered. The framework and audit duties fall only on a "large frontier developer" (revenues over $50,000,000 and AI spending of at least $1,000,000,000, measured over the preceding 36 months); transparency reports and 72-hour incident reports fall on every frontier developer. Section 9 shields every "artificial intelligence developer". Most of its areas are tied to "catastrophic risks", a term the bill defines around frontier models, but its incident-reporting area covers safety or security incidents "involving an AI model" of any kind. Second, states keep a role, but a federal one. A state attorney general who opts in may sue to collect the bill's civil penalty of "not more than $1,000,000 per violation"; if the federal Attorney General sues over the same violations, the state's case is paused (the bill's word is "stayed"). Compare what attorneys general from 24 states, the District of Columbia and American Samoa asked congressional leaders for in a letter dated 23 September 2026: "Explicit preservation of state authority by prohibiting the preemption of state laws and granting state officials full authority to enforce these federal protections." The FRONTIER Act gives them part of the second half and none of the first.

Are Democrats united on AI preemption?

No. Most of those quoted below have called for federal action on AI in some form; they divide on whether a federal rule should displace state ones.

In the Senate, Transformer reported on 9 October 2026 that Sen. Cantwell's framework for frontier AI does not address preemption. The most senior Democrats in both chambers have said AI needs action without saying, in the statements we found, whether that action should replace state law. That silence is where the narrow route lives.

What can a Democratic Congress actually change on AI?

Three things: which bills move, what the administration has to answer for, and what gets funded. The first is concrete. H.R. 9925 dies with this Congress; a reintroduced House version would most likely go to the same two committees, which the new majority chairs from 3 January. Whether a Section 9 bill gets a hearing becomes a Democratic decision. The second is oversight: hearings and subpoenas can ask where the order's promised instruments are, including the evaluation of state laws that section 4 required "Within 90 days of the date of this order", which we could not find in September. The third is money: the order runs "subject to the availability of appropriations", and Congress writes the appropriations.

Do state AI laws stay in force under a Democratic Congress?

Yes, as far as Congress goes: it will not take them away against its majority's will, but courts and the executive branch still can. Governor Newsom's office announced on 29 September 2025 that he "today signed into law Senate Bill 53, the Transparency in Frontier Artificial Intelligence Act (TFAIA)". New York's chapter amendment to the RAISE Act, S8828 (the follow-up changes agreed when the governor signed it), sets its effective date at "JANUARY 1, 2027". Colorado's SB 26-189 "repeals and reenacts those provisions with new requirements", with duties "starting January 1, 2027,".

Colorado shows the other route. As we reported in September from the court's docket, the United States joined a challenge to Colorado's law as a plaintiff on 24 April 2026, and on 27 April a minute order granted a joint motion and ordered that the defendant, Colorado's attorney general, not initiate enforcement. That standstill concerned the law before its reenactment, and I have not established how it applies to the new one. The point stands either way: federal litigation does not need Congress.

Could Democrats agree to a narrow AI preemption deal?

Yes, and the case for it is stronger than this piece's opening paragraph suggests. Four arguments, stated as their holders would state them.

First, the appetite is bipartisan. Three Democrats cosponsored the bill containing Section 9. Rep. Houchin, a Republican cosponsor, described the June draft as creating "a clear national standard for how AI models are developed while preserving the ability of states", and Rep. Trahan has said plainly that passage needs both parties.

Second, the trade is already drafted. Democrats who want binding frontier rules get them in the FRONTIER Act: licensed independent verifiers, incident reports within 72 hours, emergency orders that let the Secretary of Commerce suspend a model whose development, deployment or internal use "presents an imminent catastrophic risk", and civil penalties. The price is Section 9. A majority whose House leader, Rep. Jeffries, wants "decisive congressional action immediately" and which cannot override a veto may decide that a federal floor with preemption beats state laws under federal attack.

Third, the President may sign it. The White House wants states kept out of "AI development", and Section 9 bars state obligations on developers, not on deployers. Against that sits the strongest evidence on the other side. The same recommendations say "Congress should not create any new federal rulemaking body to regulate AI," and would instead work through existing regulators and industry-led standards; the FRONTIER Act provides for an "Under Secretary of Commerce for AI Security", appointed by the Secretary, who issues regulations. And the President's own Executive Order 14409, signed 2 June 2026, says: "Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models." That is a rule for reading that section of his order, not a veto threat. But the FRONTIER Act says a large frontier developer "may not develop, deploy, or operate a frontier model" without a current disclosure statement filed and a fee paid, and its section 8 lets the Secretary suspend a model. Whether that is a deal-breaker is the President's call, and he has not made it on the record I read.

Fourth, the executive branch may do what Congress will not. Section 6 of the order asks the FCC to consider "a Federal reporting and disclosure standard for AI models that preempts conflicting State laws", which is close to the ground Section 9 covers. If that proceeding or the task force succeeds, the congressional question matters less.

Where this leaves me: the counter-case does not rescue the broad plan. It strengthens the narrow one. After 3 January the live question is not whether Congress preempts state AI law, but whether it preempts the frontier-safety slice, permanently as the bill now reads, in exchange for federal rules.

What should readers watch after 3 January?

Whether Section 9 survives reintroduction. H.R. 9925 is a bill of the 119th Congress, so any version would be a new bill in the next. Three questions apply to any preemption text that appears, from either party:

1. Does it end? The June draft's preemption expired in December 2029, by its co-author's account. The introduced bill's does not. A sunset is one sentence, and its presence or absence is the clearest signal of what is being traded.
2. Does it name what it displaces? A clause that lists the state provisions it preempts, by citation, tells California, New York and Colorado where they stand. "New substantive obligations" leaves that to a court.
3. Who enforces? The attorneys general asked for "full authority to enforce these federal protections." An opt-in right that pauses when the federal Attorney General sues over the same violations is a smaller thing, and a reader should be able to see that difference in the text.

A Democratic Congress cannot write America's AI law alone. It can decide which preemption, if any, reaches the President's desk, and the one already written covers, on my reading, the ground its own states have legislated; whether it reaches laws already on the books turns on the word "new".

---

This analysis is by Mücahit Kaya, founder and editor-in-chief of AI Tools Police, an AI industry researcher with a public administration background. It draws on the published record, read in the original: CBS News's election projection; H.R. 9925, the FRONTIER Act and H.R. 8031, the GUARDRAILS Act at the Government Publishing Office; Executive Order 14365 and Executive Order 14409; the Congressional Record Daily Digest for 30 June 2025; the White House's legislative recommendations of March 2026; The Hill on the defence bill; the Obernolte and Trahan release and their June draft release; Tech Policy Press; Nextgov; the GUARDRAILS release; the Congressional Progressive Caucus; Bloomberg Government; the House Democratic AI commission release; States Newsroom via CT Mirror; Transformer; the attorneys general's letter; the governors' releases on SB 53 and the RAISE Act; New York S8828; Colorado SB 26-189; the Congressional Research Service's Invoking Cloture in the Senate; and the Constitution. Where the piece moves from the record to argument, it says so. The conclusions are the author's own.