Last July, the Trump administration issued its AI Action Plan. Declaring that to “maintain global leadership in artificial intelligence, America’s private sector must be unencumbered by bureaucratic red tape,” the Action Plan recommended a number of ways to eliminate any federal oversight of AI. But included in these recommendations was one that, on the surface at least, seemed… odd.
Led by the Federal Communications Commission, evaluate whether state AI regulations interfere with the agency’s ability to carry out its obligations and authorities under the Communications Act of 1934.
Given that the FCC has never claimed any authority over anything relating to artificial intelligence, and that nothing in the Communications Act appears relevant (especially since the Action Plan limits the review to 47 U.S.C. §§ 151-646, and therefore leaves out all the stuff relating to broadband), this directive to the FCC appears almost like a non-sequitur. But for those of us familiar with the FCC, and the Trump administration’s desire to preempt state laws it considers “bureaucratic red tape” (which means pretty much anything other than treats for AI companies), the reason seemed clear. The FCC is the only federal agency with an even vaguely relevant jurisdiction with preemption power.
Sure enough, in an interview at Politico’s AI summit, FCC Chairman Brendan Carr stated that the agency would soon conduct the review, and might find ways to preempt state laws, such as those adopted by California. As other states are busy exploring the regulation of AI, this serves as something of a shot across the bow – warning states not to get too excited or aggressive about AI regulation because the FCC will come and preempt you.
This raises the obvious question: Can the FCC preempt state AI laws?
No.
I caveat that in this day and age, you really can’t say anything for sure in light of the Supreme Court bending over backward to give the Trump administration whatever it wants, regardless of precedent. But if we look at the relevant precedents, this isn’t even a close call. Artificial intelligence is the quintessential “information service” as defined by 47 U.S.C. § 153(24): “offering of a capability for generating, acquiring, storing, transforming, processing, retrieving, utilizing, or making available information via telecommunications.” Indeed, now that broadband itself is definitively a Title I “information service,” AI isn’t even an information service since it is no longer offered “via telecommunications.” So, following the logic of American Library Association v. FCC, not even ancillary jurisdiction applies here.
I’ll unpack this as I look at what Chairman Carr has explicitly said so far, and other possible sources of FCC preemption authority Chairman Carr didn’t get into.
FCC Preemption Authority
We start with the general preemption of state regulation of “interstate commerce” by Article I of the Constitution (the “Commerce Clause”). This gives the federal government authority to regulate interstate commerce – and federal law, of course, preempts contrary state laws. But Congress actually has to pass a law to do this. A constitutional grant of power to Congress does not by itself preempt state law or limit a state’s power to regulate commerce within its own borders, even in ways that may have some impact on interstate commerce.
While the Supreme Court has held that states may not discriminate against out-of-state commerce (for example, laws that allowed in-state wineries to ship directly to consumers, but not out-of-state ones, were found unconstitutional), it has consistently rejected arguments that state laws that merely affect out-of-state commerce are unconstitutional – for example, by holding that California may ban the sale of pork within California from animals that were confined in a “cruel manner,” even if they are raised out of state.
So, the question becomes whether Congress actually has preempted state regulation of AI, or delegated to the FCC the authority to do so. It has not. As explained by this Congressional Research Service paper, preemption happens in three ways. First, Congress may itself expressly preempt states by simply saying “we preempt the states.” The Communications Act does do this in some places, so it is possible that Congress may have expressly preempted the states in a way the FCC can apply. Second, a regulatory scheme may be so comprehensive that Congress is considered to have implicitly preempted by “occupying the field” and leaving no room for independent state action (unsurprisingly called “field preemption”). Courts have repeatedly found that the Communications Act explicitly leaves a role for states in regulating communications within their state (see 47 U.S.C. § 152(b)), so field preemption does not apply.
Finally, Congress may delegate to a federal agency rulemaking authority, under which a federal agency may preempt states by creating a rule that makes it impossible for the state to create a law that contradicts the rule. This is called the “impossibility doctrine.” But the relevant agency must actually have the authority to make such a regulation. If the agency doesn’t have the authority to regulate something, then it doesn’t have the authority to preempt states from regulating it, either.
And now we come to my favorite irony, which I have been warning the FCC about since 2010. In the first Trump administration, then-FCC Chair Ajit Pai went to great pains to eliminate any FCC authority over broadband (let alone AI) in the 2018 “Restoring Internet Freedom Order” (RIFO) eliminating net neutrality and reclassifying broadband as a Title I service. That was why he called it the “Restoring Internet Freedom Order.” The Order went to great lengths to explain that the Federal Trade Commission and the states were the actual entities that should protect the public from any possible broadband harms – whether net neutrality, privacy violations, or whatever. Surely it was not for the FCC, the federal agency Congress created expressly to regulate communications, to insert itself into regulating broadband. Former Chairman Pai was a true-blue libertarian deliberately trying to eliminate regulatory authority as per se bad. Since the Sixth U.S. Circuit Court of Appeals reversed the 2024 Order reclassifying broadband as Title II and decided that broadband is a Title I information service as a matter of law, we now live under the FCC’s 2018 RIFO and its sweeping elimination of direct FCC authority over broadband.
But in one area, Former Chairman Pai miscalculated. The RIFO also tried to preempt states from regulating broadband. But – as every court that has considered the question since broadband became Title I has found – an agency can only preempt states where it has actual regulatory authority in the first place. Even if the FCC finds a “federal policy of non-regulation” (which the RIFO purported to find), that perceived federal policy does not convey any actual authority. And without actual authority to regulate, the FCC cannot preempt the states.
So the FCC can only preempt states where Congress has given it explicit authority over information services – although, as I indicated above, artificial intelligence may not even be an information service at this point. But the FCC also has something called “ancillary authority.” Where Congress has given the FCC authority to regulate something, the FCC may extend its power to something adjacent where it finds it necessary to do so to fulfil the direct command of Congress. So although the FCC has not officially classified facilities-based Voice Over IP services as a Title II service, it has used its ancillary authority to preempt state regulation of VOIP.
With all this in mind, let’s look at Chairman Carr’s claim of authority and some other possible claims.
Chairman Carr’s Claim of Authority
Here is what Chairman Carr is reported to have said: “We do have some authorities under Section 253 of the Communications Act. Effectively, if a state or local law is prohibiting the deployment of ‘modern infrastructure,’ then the FCC has authorities to step in there.”
However, this is not what the statute actually says. The relevant provision says that no state law or regulation “may prohibit or have the effect of prohibiting the ability of any entity to provide any interstate or intrastate telecommunications service.” (Emphasis added) That’s not “modern infrastructure,” even if we could reasonably conclude that AI counts as infrastructure, it is not a telecommunications service. The FCC has implicitly acknowledged this limitation in the proposed Notice of Inquiry that Chairman Carr referenced in the article by asking to what extent broadband services are “comingled” with telecommunications services – since the FCC can regulate (and preempt states) where the same network is used for both telecommunications and information services. But AI does not offer telecommunications services, nor does the broadband network on which it rides. Broadband is an information service, which is why Chairman Carr’s effort to preempt states to promote broadband needs to find a telecommunications service in there somewhere. And AI is one step further removed since it doesn’t touch a telecommunications service anywhere.
What about ancillary authority? As I alluded to above, an information service has to be delivered “via telecommunications.” But broadband, based on the Sixth Circuit Court of Appeals decision, is not a telecommunications service. It is an information service. While courts have said that broadband offers “communications” (and therefore falls in the broad general jurisdiction of the FCC), it is not telecommunications (subject to Title II). AI is therefore not “delivered by telecommunications” as required by the definition of information services.
But even if we assume that AI is an information service, that doesn’t really help the ancillary authority issue here. In ALA v. FCC, the D.C. Circuit Court of Appeals found that the FCC’s jurisdiction could not extend to matters that occurred before or after the actual transmission of services. The FCC therefore lacked ancillary authority to impose digital-rights-management on broadcast TV devices. Similarly, in MPAA v. FCC, the D.C. Circuit found that the FCC’s general authority over broadcast did not give it authority to require licensees to provide video description service because that would be requiring broadcasters to create specific content – which Congress had not authorized and was not the same as general obligations such as an obligation to create local programming generally. Nor could video description service be considered “reasonably ancillary” to closed captioning for the hearing impaired and the general policy of making television accessible to all.
I therefore don’t see any way for the FCC to legally claim authority to regulate, and therefore preempt, any state laws about AI – either through Section 253 directly or ancillary authority. But are there other sources of authority? Here is a quick review of what I can think of and none of these options seem promising.
General authority provisions. The FCC has two fairly broad provisions for general authority. 47 U.S.C. § 201(b) prohibits any “unjust and unreasonable practices” in the provision of telecommunications services. 47 U.S.C. § 303 gives the FCC fairly broad authority to regulate wireless services. These broad authorities have been used by the FCC to, at times, preempt contrary state authority. But we aren’t talking about telecommunications, so Section 201 doesn’t apply. It’s also hard to see what wireless applications state regulations might interfere with that the FCC could preempt. We’re not talking about antenna sitings.
The FCC has used certain other provisions that control local cable franchising (47 U.S.C. §§ 541, 42, 44) and approval of commercial mobile radio services (47 U.S.C. § 332(c)(3) & (7)) that they have “interpreted” the statutory limitations to preempt states and localities or obligate them to do things (for example, interpreting “reasonable time” to be a specific number of days). Again, there aren’t any provisions relevant to information services.
Specific provisions. Some provisions of the Communications Act directly address information services. But none of them really seem relevant either as a source of direct authority or ancillary authority. The most likely relevant is 47 U.S.C. § 1302, which requires the FCC to take steps if it finds “advanced telecommunications capability” (which has basically come to mean broadband) is not being deployed to all Americans in a timely manner. But there are several problems with trying to use this. First and foremost, Former FCC Chairman Ajit Pai disemboweled it in the Restoring Internet Freedom Order. Whereas the FCC under then-Chairman Julius Genachowski found that Section 1302 (formerly known as Section 706, since it was Section 706 of the Telecommunications Act of 1996) provided the FCC with actual authority to do things, the RIFO reversed that and found that Section 1302 was simply policy and did not give the FCC any new authority. (As I mentioned above, Ajit “Regulation Weed Whacker” Pai was determined to eliminate the FCC’s regulatory authority over everything but robocalls.) Since the Sixth Circuit set aside the FCC’s 2024 Internet Classification Order that had reversed former Chairman Pai’s reversal, we default back to “Section 1302 is policy not authority” – at least until the FCC changes its mind again and a court affirms such a decision as the “best reading” of the statute. But even if Chairman Carr decided to reverse Pai and reinstate Sec. 1302 as a source of authority (and a reviewing court upheld Carr’s decision now that the FCC no longer gets deference)(and that got upheld), I don’t see how making sure broadband gets deployed in a timely fashion to all Americans has anything to do with AI, even with the most generous application of ancillary authority I can imagine.
Additionally, Section 1302 is no longer considered part of the Communications Act and therefore, arguably, doesn’t get any authority – ancillary or otherwise. The Action Plan defines the Communications Act as Sections 151-646, which is an argument that gained traction during the rulemaking over digital discrimination (codified at 47 U.S.C. § 1754) and, I suppose, means that the internet is not part of the Communications Act. The absurdity of this position is that Section 1302 used to be codified as a note to Section 153(24) (definition of information services). So from 1996 to 2008, what we now call Section 1302 was part of the Communications Act, but when Congress expanded it in 2008, it got recodified as Section 1302 and was suddenly not part of the Communications Act. But I digress.
Conclusion
Try as I might, I simply do not see how the FCC can assert jurisdiction over AI that would give it the authority to preempt state regulation of AI. Heck, at this point, I don’t see how any application could be an information service, since applications delivered over the internet are no longer delivered by “telecommunications” but by information services. (The FCC maintains that “telecommunications” and “information service” are mutually exclusive categories.) Perhaps someone more clever than I will suggest something in the Build America: Eliminating Barriers to Wireline Deployment Notice of Inquiry the FCC will vote on at the September meeting. Until then, I remain highly skeptical that the FCC has any authority here.