Read this post on Neil’s Substack: Getting Out of Control.
On Thursday night President Trump signed an executive order titled “Ensuring a National Policy Framework for Artificial Intelligence.” Here’s some signing comments by Trump and commentary / explanation by Crypto and AI Czar David Sacks, who drove this effort:
The EO takes seven actions:
- Sets the policy of the U.S. as to “sustain and enhance the United States’ global AI dominance through a minimally burdensome national policy framework for AI.” (Much of the order applies only to state AI laws that violate this policy. I’ll call them “conflicting State AI laws” for short.)
- Creates a Dept. of Justice Task Force to challenge conflicting State AI laws.
- Tasks the Dept. of Commerce with identifying existing conflicting State AI laws and publishing a report.
- Requires Commerce and agencies to make certain kinds of funding conditional on whether the state has or enforces conflicting AI laws.
- Directs the Federal Communications Commission to begin a proceeding on whether it should require AI model reporting that preempts conflicting State AI laws.
- Requires the Federal Trade Commission to issue a policy statement detailing when conflicting State AI laws are preempted by the FTC Act’s prohibition on deceptive acts or practices.
- Directs presidential advisors to prepare draft federal AI legislation that preempts conflicting State AI laws, with no preemption for four buckets of state laws, including “child safety protections.”
FIRST THINGS FIRST: If you are reporting on the EO or arguing about it online, I implore you to READ IT YOURSELF. It’s only 1400 words, and it’s clearly written with little legal jargon. You’ll save yourself the potential embarrassment of repeating incorrect talking points from people who are misrepresenting it out of ignorance or malice.
But there are somethings that might not be obvious to everyone from reading it. Here are my key takeaways, Section by Section. You should think of these as the key things people might fight over in the EO, or key things they might ignore as inconvenient to their position.
SEC. 1 — PURPOSE
What it does: Sets forth the purpose of the EO.What you should know: This is important: The President clearly intends the EO to serve as a stopgap against the worst state AI laws until Congress does the necessary work of establishing a minimally burdensome national standard that protects kids, prevents censorship, respects copyrights, and safeguards communities. This is not a permanent “fix.” Importantly, the EO CREATES NO NEW PREEMPTION. The EO clearly and properly recognizes that the executive branch cannot do that. The statements in the video above as well as the text of the EO drive home that Congress must act. And when it does act, Congress must preserve an important roles for states, while recognizing that the federal government must lead on this nationally important technology.
SEC. 2 — POLICY
What it does: The EO sets as the policy of the United States “to sustain and enhance the United States’ global AI dominance through a minimally burdensome national policy framework for AI.”
What you should know: This is based.
SEC. 3 — CREATION OF A DOJ AI LITIGATION TASK FORCE
What it does: Establishes a Task Force at the DOJ with tho sole purpose of suing states for conflicting AI laws.
What you should know: Two observations. First, DOJ doesn’t need an EO to challenge illegal and unconstitutional state laws. They have that authority now. But this creates a institutional structure who will be held responsible for doing so.
Second, as I already noted, this does not ban any state laws that were legal before the EA was passed. It doesn’t preempt state laws. DOJ will need to persuade a court that every law challenged is unlawful under current law.
Of course, litigation imposes costs on defendants even if they win, so this task force could have an overall chilling effect on state AI legislation. That’s the point. Up until now, states faced little cost of any kind for imposing vague, unworkable, and extraterritorial restrictions on AI developers, deployers, and users. Now they’ll at least have a reason to think twice.
SEC. 4 — EVALUATION OF STATE AI LAWS
What it does: The Secretary of Commerce must publish an evaluation of existing conflicting State AI laws, and identify which laws should be referred to the Section 3 Task Force.
What you should know: The EO singles out for scrutiny laws that implicate speech, including those that “require AI models to alter their truthful outputs” or those that violate the Constitution by requiring disclosures or reports by AI developers or deployers. First Amendment lawyers, start your engines — there are really interesting questions here.
Echoing past language from various congressional measures on preemption, the Secretary is also permitted to “identify State laws that promote AI innovation…” This could inform the Sec. 8(b)(iv) “other topics” carveouts from preemption that will be in the White House’s recommended legislation.
I suspect there are going to be a lot of state Governors and other stakeholders seeking to meet with the relevant Commerce staff to lobby for their various state laws. I can already imagine some of the arguments they’ll make.
SEC. 5 — RESTRICTIONS ON STATE FUNDING
What it does: Substantively, this is the most complex requirement of the EO. It obligates Commerce to issue a Policy Notice specifying that states with “onerous AI laws” as identified in the Sec. 4 report discussed above or challenged by the Sec. 3 Task Force “are ineligible for non-deployment funds” from the Broadband Equity Access and Deployment (BEAD) Program, “to the maximum extent allowed by Federal law.” This section also directs other “executive departments and agencies” to determine whether they can condition any discretionary grants on states not passing or enforcing conflicting AI laws.What you should know: I am not sure how large a bucket of BEAD money this involves (one of my telecom law buddies probably knows) or to what extent federal law would permit these kinds of conditions. However, this does strike me as one of the more legally risky areas of the EO, because there are large private telecommunications companies who would be receiving this money from the states and who may have the incentive and means to sue to challenge any such conditions, if they are applied aggressively. As for the other agencies’ $$$, it is even less clear how much money this affects — it’s the sort of thing that probably would be hard for even the White House to determine independently. That’s why the agencies are tasked with it. But I suspect this isn’t a massive amount of money. On top of that, most agencies probably don’t want to mess with their existing programs and may resent this extra work. Institutional incentives lean toward agencies minimizing the amount affected. As such, I expect this to be a relatively low-impact provision.
SEC. 6 — PREEMPTIVE FEDERAL REPORTING REQUIREMENT
What it does: This section requires the Federal Communications Commission to start a proceeding asking whether it should adopt a Federal reporting and disclosure standard for AI models that preempts conflicting State laws.
What you should know: Note that this doesn’t require the FCC to actually adopt such a provision. It requires what is known as a “Notice of Inquiry” or “NOI”, which is what agencies sometimes do before they start a rulemaking to ask whether they actually should start a rulemaking.
My own initial view is that the FCC would be a strange place to house such AI reporting and disclosure requirements, and I have questions about the FCC’s legal authority to do this. But I look forward to digging in and commenting on the forthcoming NOI.
SEC. 7 — FTC UDAP PREEMPTION
What it does: This section directs the Federal Trade Commission to issue a policy statement identifying situations in which a State requirement to “alter[] truthful outputs of AI models” is preempted by the FTC Act Section 5’s “Unfair and Deceptive Acts or Practices” (UDAP) authority.
What you should know: This section is fascinating to me because during my time at the Federal Trade Commission I dealt with many dozens of cases involving the FTC’s UDAP authority. I have never seen it applied like this, but it doesn’t strike me as obviously wrong. The theory seems to be that if a State law requires a company to lie, but Section 5 prohibits a company from lying, those laws are in direct conflict and therefore Section 5 preempts the law. I guess this Policy Statement would be used in court by companies defending themselves against such laws?
I want to think more about this, including what it could mean for other state laws that arguably require “lying.” For example, California’s cancer labeling requirement probably wouldn’t be substantiated under typical FTC standards. There are a bunch of green labeling / environmental disclosure requirements that similarly probably require companies to bend the truth, or at least not fully represent its nuance.
Also, the deception statement doesn’t mean every false statement is a violation. To be deceptive under section 5 a false statement has to be material to a customer, meaning they would have acted differently if told the truth. Does that mean the state AI law is only preempted where the required deception would be material?
Anyhow, very early thoughts on this — I will be writing more.
SEC. 8. — LEGISLATIVE RECOMMENDATION
What it does: Consistent with the stop-gap nature of the EO, this section jointly tasks the Special Advisor for AI and Crypto and the Assistant to the President for Science and Technology with preparing a legislative recommendation “establishing a uniform Federal policy framework” that preempts conflicting state AI laws.
What you should know: New to this version (wasn’t in the draft) are four areas carved out from what the recommendation may recommend preemption. The recommendation will not include preemption of :
- child safety protections;
- AI compute and data center infrastructure “other than generally applicable permitting reforms”;
- State government procurement and use of AI; and
- “other topics as shall be determined”
That last bucket could include state AI laws that promote AI development or deployment. The second, infrastructure carveout is also interesting: it appears to preserve the legislative draft’s ability to recommend preempting certain state permitting practices.
These carveouts make crystal clear what supporters of various measures to contain state laws, all the way back to the July moratorium fight, had attempted to explain: there are definitely areas where states have an important role and should not be preempted.
SEC. 9 — GENERAL PROVISIONS
This is just the usual Executive Order boilerplate.
FINAL THOUGHTS
This EO is not a silver bullet, and it doesn’t pretend to be one. It does not magically wipe away state AI laws, nor could it. What it does instead is more subtle and more realistic. It raises the cost of the worst forms of state AI regulation, creates institutional pressure to test their legality, and clearly signals that the status quo of fifty competing AI regimes is unacceptable for a technology that operates at national and global scale.
Most importantly, it frames the executive branch’s role correctly: as a bridge to legislation, not a substitute for it. The hard work now shifts to Congress, where the real question is not whether there should be a national AI framework, but how it can be drawn to ensure continued American AI dominance, including by preempting overreaching state laws while preserving state authority where it makes sense.
In that sense, the EO succeeds if it does one thing above all else: it forces the debate out of abstraction and into concrete legal, institutional, and political tradeoffs. That debate is long overdue.