What the leaked AI executive order tells us about the Big Tech power grab

Hello and welcome back to Regulator. It’s been a very long two weeks away from your inboxes, but luckily for us, Big Tech and Big Government did not stop fighting. In fact, it’s gotten even spicier. Let’s get into it.

Last week, I was following up on several rumors that Donald Trump would sign an executive order that would fulfill a longstanding goal of the AI industry: legal preemption that would prevent states from passing their own AI laws. Mostly, I was calling sources trying to get a sense of how the Trump administration planned to approach it: Which agency would be spearheading it? What legal arguments would they use? How would it interact with Congress, which was trying to pass a similar moratorium in the National Defense Authorization Act?

And then I got a copy of the draft order itself — possibly a sign that someone in the administration deeply, deeply loathes David Sacks, Trump’s Special Advisor on AI and Crypto. Even though he’s not a permanent government employee — he is, in fact, a billionaire tech venture capitalist with a provisional employment status similar to the one Elon Musk previously held — Sacks has become deeply influential in setting the administration’s AI and crypto policies. (Just look at Trump’s recent statements about federal AI preemption.)

Leaks rarely come out of Donald Trump’s White House these days, especially compared to his previous term. Back then, everyone in the Trump administration was trying to undermine everyone else on a regular basis, dropping juicy, scandalous tidbits to their favored reporters on a near-hourly basis. In that first term, the career, norm-following officials of the federal bureaucracy were anxiously texting journalists about chaos they’d been preventing. Conservative media figures were eagerly telling reporters about their late-night phone calls with Trump. And the Fifth Avenue New Yorkers were fighting the Steve Bannon–led populists, who were fighting the establishment Republicans, who were fighting the Democrats, who were fighting each other.

There are fewer leaks this time around, primarily because the administration is built around blind loyalty to the president. DOGE gutted the bureaucracy and Trump hired people who would always tell him yes. But on that rare occasion that an actual document leaks out of the administration, it’s a sign that someone has carefully weighed the risks of undermining their enemy, considered the cost of doing so, and decided they hated them enough to do it anyway.

Which brings us to this draft executive order. If it had been signed, it may not have been an actual ban on state AI laws, but it would have given the executive branch the power to strongly dissuade states from writing or enforcing their own laws. Below the break, I walk through the policy implications of the order with Charlie Bullock, a senior research fellow at the Institute for Law and AI, who pointed out all the ways that the government would be able to punish the states for trying to regulate AI: suing them, withholding billions of dollars in federal funding, and hitting them with FTC fines.

While the order itself might eventually be found illegal, he noted, the order would make it painful for a state to fight back: “A state that really needs broadband funding, for example, could say, It might take a long time for us to get our funding. Even if it can win a court case to make them give us that funding eventually, it would take a long time. States might be incentivized not to pass legislation contrary to the policy of the order.”

But politically, the second Trump administration takes an approach to executive orders that’s more shoot first, ask questions about its legality later. Other equally overbroad orders haven’t leaked prior to their signing, and there was nothing spiritually different about this executive order, save for who had been empowered most: every directive included language that required the government to consult the Special Advisor for AI and Crypto, who happens to be a certain tech billionaire from the private sector from outside the political world. And that power play was clearly enough for someone to break Trumpworld omerta.

We’re going to go into the parameters of the power grab below, but before we do, here’s the latest from The Verge:

“I suspect that if it’s effective, it’ll probably be by having a chilling effect on state legislation”

This interview has been edited for clarity.

Tina Nguyen: This order is supposed to, somehow, implement a moratorium on state AI regulation. We’ve seen that try to take place via legislation that’s clearly failed, and probably going to fail again. How effective would this order as it stands be in actually making a moratorium happen?

Charlie Bullock: In my opinion, this order cannot make a moratorium happen — literally. It’s an executive order. An executive order is not congressional legislation. Executive orders can do a number of things, but mostly what they do is they announce the policy, goals, and opinions of the executive, of the president, and direct executive branch agencies to take actions.

So the way that this executive order tries to implement preemption, you could say, is that it tells the Department of Justice to establish a task force to sue states over their AI laws. An executive order could do that, because the Department of Justice is within the executive branch.

What it cannot do is announce, Okay, state AI laws are now preempted. An executive order cannot unilaterally override state laws. It can announce executive actions that are going to happen that might, in effect, stop states from promulgating new AI legislation or it could, in theory, invalidate existing AI legislation, but it cannot just override it.

Now, as for how effective it will be in what it tries to do, it’s difficult to predict how the future will go, especially in situations that are legally complex. I suspect that if it’s effective, it’ll probably be by having a chilling effect on state legislation. For example, section 5 of the order regards restrictions on state funding. They’re announcing that they’re going to attempt to withhold various federal grant funds for the states that would otherwise get them, if those states have so-called “onerous AI laws” — laws that are contrary to the policy positions expressed in the order. That does not directly preempt state laws or get rid of any state laws. But a state that really needs broadband funding, for example, could say, It might take a long time for us to get our funding. Even if it can win a court case to make them give us that funding eventually, it would take a long time. States might be incentivized not to pass legislation contrary to the policy of the order.

So framing it more as a chilling effect rather than “no more state AI laws” is the correct way of looking at this.

Yes. In theory, this AI litigation task force could find some great legal argument and go to court and get a lot of state AI laws knocked out. But the arguments that are specifically mentioned in the executive order are pretty weak, in my opinion, and unlikely to succeed in court if states fight back against them. I think it’s likely that states like California have a strong interest in regulating AI and are often in opposition to the Trump administration and have political incentives to oppose the Trump administration in various ways. If they fight it in court, I think that at least on the legal theories that are mentioned in the executive order, that task force would not succeed.

The FCC’s inclusion seems to have sparked a new discussion of whether telecom policy has any influence on AI law.

So section 6 says that the chairman of the FCC shall, in consultation with David Sacks, initiate a proceeding to determine whether to adopt federal reporting and disclosure standards that preempts conflicting state laws. Now, that’s nothing illegal because all you’re doing is initiating a proceeding for determining whether you do something. But if the FCC actually tried to promulgate a federal reporting disclosure standard for AI models that preempted existing state laws, they would have to have some legal authority, probably congressionally granted, to do that.

As far as I know, and as far as people with experience in telecom law have written about this know, the FCC just does not have authority to regulate it. It’s not clear exactly what legal theory they’re relying on there. We have some past statements by the FCC chair and various people involved in the FCC that sort of indicate that maybe they think they do have some sort of regulatory authority over AI, but as far as I know, they do not. So that means that they’re not legally going to be able to promulgate that standard and have it be effective, or conflict with state laws for being contrary to that standard.

Can we talk about what the FTC is being empowered to do here? It sounds like they’ve been granted an ideological, anti-woke enforcement mechanism, but you’re able to describe it in a much more informed manner.

The FTC Act prohibits unfair deceptive acts or practices concerning commerce, and it’s a consumer protection provision, essentially. The FTC can do enforcement to prevent things like scams, things like false advertising and stuff like that. I’m really not an expert on FTC Act stuff or on algorithm discrimination laws, but as far as I know, this is the first time anyone’s ever attempted to do anything like this to preempt state laws for this kind of reason. They’re basically saying: algorithm discrimination laws like Colorado’s, and presumably other so-called woke AI laws, are deceptive because they require models to say untruthful outputs in some way, or something like that. And because they require alterations to make untruthful outputs in AI models, they’re preempted by the FTC Act’s prohibition on engaging in deceptive acts concerning commerce.

Again, I’m not an expert on this area, so I don’t know how plausible that argument is. I’m not gonna speculate on whether it’ll succeed in court, if it’s challenged or something. But I’ve never seen anything like it before.

The most power in this executive order is given to the Secretary of Commerce. And it looks like it’s two things that he’s being directed to do: in section 4, they want to evaluate any state laws that are inconsistent with what Trump wants. And then section five is determining which states could get their BEAD [the Broadband Equity Access and Deployment program] funding pulled. Is that a correct way of describing it, and how easily could those two be stitched together into something quasi-legal?

So section 4 has no real, substantial legal effect on states on its own, but it’s asking the Commerce Department, in consultation with a bunch of other figures in the White House, to publish a list identifying the bad laws, the “onerous laws,” the laws that conflict with the policy set out in the EO.

That list is going to inform a lot of the other substantive parts of the order. Section 5’s restrictions of state funding: which states have laws that are on the bad list? Section 3: okay, we’re going to sue states for having bad laws, which laws are bad? The Task Force is going to decide, but they can also look at the bad list and that can tell them. Likewise, there’s no requirement that the FCC try to preempt the laws from section 4, but they might also try to preempt other laws.

Is there one detail that’s been overlooked that you need to highlight in this EO?

So you know how the section 5 restrictions on state funding is split into two subsections? Section 5(a) is instructing Commerce to withhold all the non-deployment funding from BEAD from states with laws identified as onerous. It’s about 42.45 billion that the states are supposed to get. And then section 5(b) is really interesting. It instructs all other agencies to review all of their discretionary grants, and see if any of them can be withheld from states with “onerous AI laws.”

So not just BEAD — anything could fall into this bucket?

All federal discretionary grant funds, and that’s a ton of money. Hundreds of billions of dollars. There was recently a different legal fight over highway funding: the way that the federal government pays for interstate maintenance is that it gives grant money to states, and then the states use it to fix roads or whatever. There was recently an attempt by the Trump administration to add a condition to all the Department of Transportation grants, saying: you have to help the federal government enforce immigration law if you take this money. And states sued, and they were successful, at least at the district court level, the lowest court level. We’ll see what happens on appeal, but they were successful in saying, this is unconstitutional, and it’s unlawful. You can’t impose this requirement on us.

Essentially what the [AI draft order] is saying is, Those highway grants, any education grants, all discretionary funding, in a ton of areas with the federal government gives money to the states, we’re gonna look at all of them and see if we can legally withhold any of them from you if you have bad AI laws. So there’s potentially tons of money that a lot of states want, and even if they succeed in suing, it could take a while for them to get that money. Even the delay in receiving funding could be impactful.

I may have been out for two weeks, but I wasn’t living in a news cave. In fact, I was eagerly mainlining two huge stories: New York City mayor-elect Zohran Mamdani meeting Trump in the Oval Office, and Rep. Marjorie Taylor Greene (R-GA) breaking from Trump and announcing she would leave Congress in January.

I also like vintage memes.