Preemption Isn’t Governance. We Need a Federal Framework for AI.

Last December, President Trump signed an executive order (EO) directing federal agencies to identify and preempt state artificial intelligence laws that conflict with the administration’s goal of maintaining American AI dominance. The order echoes many of the usual industry talking points on state-level AI regulation – that a patchwork of burdensome rules threatens innovation, increases compliance costs, and hurts small developers. Some of these claims have real merit! However, none of them justifies signing away states’ rights to responsive, active governance of an ever-evolving technology. What the Order Actually Does The AI EO contains a variety of different mechanisms designed to preempt state laws. However, executive orders don’t preempt state law on their own – they merely direct a federal agency or encourage Congress to act – and many of these mechanisms face real political or legal hurdles. For one, the EO instructs federal agencies to use every available tool, including litigation, regulation, federal enforcement, and legislative proposals, to block state AI laws deemed inconsistent with federal policy. In practice, that means agencies could pressure states through lawsuits, aggressive preemption arguments in court, and behind-the-scenes lobbying for federal language that overrules any state standards. Beyond wiping out existing laws, it could also likely chill enforcement against AI companies or users and deter state legislatures from even legislating on areas of general applicability like privacy, consumer protection, and civil rights. Additionally, it attempts to leverage the Broadband Equity, Access and Deployment (BEAD) program nondeployment funds – i.e., any money left over of the $42 billion appropriated for rural broadband after deployment is accounted for – as a cudgel against states that enact AI regulations. Given that the BEAD program has been reconfigured to favor cheaper broadband technologies, there’s a whole lot of money still left in the program – potentially half of the fund, or about $20 billion. So tl;dr, the administration is saying: states can comply with these federal AI priorities or lose billions in broadband funds – which can be used for things like network modernization, workforce development, and digital adoption efforts. However, while Congress explicitly appropriated BEAD funds for broadband deployment, AI isn’t mentioned in the statute once. So even under a generous reading, courts would likely side with states that contested the withholding of funds on “achieving AI dominance” grounds. And we’ve already seen some pushback to the administration’s attempts to unilaterally redirect congressionally appropriated broadband funds – the National Digital Inclusion Alliance filed a lawsuit after the administration terminated the Digital Equity Act’s competitive grant program, a companion to BEAD in the same infrastructure law. Lastly, the EO directs the chair of the Federal Communications Commission to “determine whether to adopt a federal reporting and disclosure standard for AI models that preempts conflicting State laws,” the FCC will likely conclude… well, that they can’t do much of anything. The FCC’s jurisdiction is communications and it has no substantive statutory authority over AI. The current FCC doesn’t even want to claim its clear Title II authority over internet service, so any attempt to stretch its authority to somehow preempt state AI laws would face serious legal challenges. So while the White House can advance a legislative framework or prioritize preemption arguments in litigation, it’s up to Congress to pass laws with preemptive effect. The Congressional State of Play Earlier this year, House Republicans passed a ten-year moratorium on state AI regulation as part of the reconciliation package. But then, after immense backlash from a diverse cast of stakeholders including: governors, state attorneys general, child safety groups, consumer advocacy groups like Public Knowledge, AI safety researchers, and plenty of everyday Americans – the Senate voted 99-1 to strip it out. Many states have laws regulating AI on the books, including Colorado, Utah, Tennessee, New York, and California. Between the messiness of openly bucking with their own state legislatures and the 10 year preemption window (remember: ChatGPT wasn’t even around 10 years ago!), it proved to be a difficult sell for many Republicans. However, even after that overwhelming, bipartisan rebuke, the effort retains a strong white whale allure for many members of Congress. Sen. Ted Cruz has said the moratorium “will return,” and as the chair of the Senate Commerce Committee, he holds unique power to keep the idea alive. The most recent effort has been an attempt to insert a preemption provision into the must-pass National Defense Authorization Act – a last-minute move that suggested proponents knew it couldn’t pass through regular order. That failed too. Democrats have diverse perspectives on AI, ranging from more techno-optimistic to skeptical, but nobody has expressed support for broad preemption without a federal regulatory framework in place – and Democratic leadership has actively opposed broad preemption proposals. Meanwhile, Republicans are divided. Some see any regulation as an innovation killer, but others like Sen. Marsha Blackburn of Tennessee have states’ rights or child safety concerns. Any preemption bill that needs 60 Senate votes (which it will, absent another reconciliation vehicle) will simply have to offer up a meaningful compromise Democrats can accept. There’s evidence that a handful of pro-AI Republicans are starting to internalize that, but such a proposal still does not appear to be forthcoming. But if AI is indeed existentially threatened by the myriad state AI bills being passed, that should encourage some kind of bargaining – especially given that the House could flip this year, meaning the legislative landscape could get far worse for preemption supporters. How Well-Designed Preemption Could Serve the Public Interest Public Knowledge’s position is straightforward: it sometimes makes sense to pursue federal regulation over state regulation, and thus federal preemption of state AI laws may be warranted – but only on a 1:1 basis. For anything we preempt, there must be an adequate federal framework addressing that specific thing. You may ask why we’d ever want to take a matter out of states’ control, but consider the areas where an AI application’s harm crosses borders by default, or where you want national-level cohesion (such as with laws governing trade or national security). Does it make sense for Kansas to have different laws on AI and bioweapons than Missouri or California? Not really. This is how Congress has handled preemption in other technology contexts. For instance, the Federal Food, Drug, and Cosmetic Act preempts state medical device requirements, but it does so as part of a comprehensive federal regulatory regime with robust FDA oversight. Preemption always corresponded with the scope and type of federal action by the relevant regulator. But the reconciliation moratorium would have preempted virtually any state law regulating AI – a definition so broad it would encompass many consumer protection, civil rights, privacy, and health and safety laws – and replaced it with literally nothing. The same logic applies to the executive order. Directing agencies to challenge state AI laws while offering no affirmative federal protections for consumers, workers, or civil rights is just an abdication of leadership, and people are right to be anxious about the potential ramifications. If Congress wants to move on AI preemption, a reasonable deal could simply instate narrow preemption of specific, defined categories of law and pair it with strong federal action in that same category. For instance, they could implement federal frontier model transparency mandates while also pairing that with preemption of the same issue at the state level. And any package must be explicit about what is not preempted as well, such as state consumer protection laws of general applicability and civil rights laws, to discourage actors from gaming (or litigating against) the system. This approach allows states to remain nimble amid congressional gridlock. If Congress truly identifies a national-scale priority and can get itself together to act affirmatively, then preemption can be a tool for creating clarity and certainty. But the regulation-impedes-innovation mindset currently underpinning the Trump and congressional Republican preemption effort is entirely backwards: giving the public a hand in shaping the technology and showing the government has a hand on the wheel can build the trust to accelerate adoption. By contrast, broad deregulatory preemption could further polarize an already-skeptical public. The administration is right that a patchwork of state AI frameworks creates real compliance challenges, but the answer to that problem is not to forgo meaningful federal leadership. States need time to discern the risks and opportunities of this constantly evolving technology, and they must remain free to engage in an iterative process that could help inform and strengthen our federal response. The administration’s failure to secure a broad state AI law moratorium is a useful reminder that unambiguously bad policy breeds powerful resistance. More sweeping preemption attempts are coming – including through the pending legislative draft responding to Trump’s AI framework – but we have a bipartisan network ready to mobilize against them. Real AI governance means Congress doing the hard work of negotiating what protections Americans actually need. That process will necessarily take a while; it cannot be cobbled together overnight for expediency’s sake. Public Knowledge will continue to engage in that process, and to push for the same intentional approach that’s guided past innovation.