How Trump v. Slaughter Strengthens the Case for Third-Party AI Regulation

In the world of artificial intelligence (AI) governance, important developments don’t always come with the words “artificial intelligence” attached to them. That was the case with last month’s decision in Trump v. Slaughter, in which the Supreme Court overruled a New Deal-era precedent and held that the president has the power to fire a federal trade commissioner to advance his policy priorities, despite a law limiting such removals. The case is sure to have widespread consequences for the federal government across the board, as removal protections have been part of the structure and politics of many independent agencies for decades. But it is likely to be particularly significant for a key challenge facing the country: developing effective, accountable governance of AI, which is both increasingly capable and increasingly integrated into the economy.

Trump v. Slaughter takes off the table one of the most popular proposals for AI governance: an independent federal agency, run by a set of commissioners who cannot be removed at the president’s whim. In doing so, it leaves proponents of AI regulation with fewer options, forcing them to choose between direct regulation by a federal agency without political independence from the president or the greater independence offered by regimes that incorporate private third-party regulators. Meanwhile, the recent actions of the federal government provide many lessons for this debate—lessons that make additional layers of political insulation look more and more appealing. The result may be that, although the Slaughter majority aimed to restore presidential control over policy, the best response in the world of AI governance ends up being to diffuse some control to non-state actors while maintaining an ultimate accountability check in a federal oversight agency.

The Case for Third-Party Regulation Before Slaughter

Over the past few years, there has been a growing chorus of arguments in favor of incorporating third-party regulation into AI governance in some way. This could take a few forms: current significant visions include a variety of elements, such as a regime in which independent auditors assess companies’ own safety and security claims, or one in which the government sets standards and auditors evaluate companies’ compliance with those standards. Third-party auditing could be opt-in (in exchange, say, for liability limitations) or mandatory; auditors could compete in what Gillian Hadfield and Jack Clark call a “regulatory market” or be funded by the government.

Why rely on these third parties at all, instead of just directly having the government regulate and audit AI developers? The standard arguments have focused primarily on two areas: speed and human capital. Regulating AI requires a system that can pay large amounts of money to hire scarce technical talent and that can pivot rapidly to address new types of concerns. Historically, government agencies struggle in both of those areas, operating on civil service pay scales and multiyear rulemaking cycles. It might be possible to adapt the federal agency model to fit the occasion, but that also might run afoul of the Supreme Court—as happened, for instance, with the Court’s rejection of the Consumer Financial Protection Bureau’s independent solo director model.

The case for third-party regulation is not a slam dunk. In particular, third-party regulation has serious principal-agent problems. To take one famous example, Esther Duflo and her co-authors found in a study in the Indian state of Gujarat that auditors chosen and paid by the firms they audited systematically underreported pollution. (In contrast, when auditors were instead randomly assigned, paid fixed fees from a common pool, and monitored better, their reporting increased in accuracy and pollution actually fell.) The U.S. financial system’s reliance on third-party credit rating agencies contributed to the financial crisis of 2008. Nongovernmental AI auditors raise serious concerns with safetywashing, where an ineffective auditor’s seal of approval helps developers appear safer than they really are. Some of these problems are well understood. But understanding them is not the same as solving them, especially because politics and capture will be at work during the regulatory design process itself.

Nonetheless, a strong role for third parties in AI regulation is starting to gain traction. Just this week, Governor JB Pritzker signed legislation making Illinois the first state to require annual independent third-party audits of frontier AI developers’ safety practices, and Connecticut’s new AI law creates a pilot program for state-approved independent verification organizations to assess AI systems. These are still the early days of third-party involvement in AI governance. But these are still the early days of AI governance, period, and these recent bills are a testament to what policymakers think is a plausible direction.

2026 and the Perils of Executive Discretion

On top of the more common speed-and-flexibility arguments in favor of third-party regulation, the events of 2026 so far have raised the profile of another factor: political insulation.

The past few months have highlighted some of the more negative possibilities that can come with direct involvement of political officials in the regulation of AI companies. First, in late February came Anthropic’s escapade with the Department of Defense. After Anthropic refused to agree to certain contract terms with the department, Secretary of Defense Pete Hegseth went nuclear, designating Anthropic a supply chain risk (an act typically reserved for foreign entities) and announcing that no company that does business with the U.S. military could “conduct any commercial activity with Anthropic.” These moves struck many as unlawful right off the bat and were enjoined by a federal court.

Then, in mid-June, came the export controls issue with Fable and Mythos. Three days after Anthropic launched its most capable models yet, and after two months in which the company had engaged in a limited release and briefed and consulted the government, Commerce Secretary Howard Lutnick ordered Anthropic to suspend access to Fable 5 and Mythos 5 by any foreign national anywhere in the world. In practice, that forced Anthropic to cut off access to the models entirely. The Friday-evening directive, by Anthropic’s account, did not detail the underlying security concern. Within a few days, a large group of cybersecurity experts spoke out against the government’s actions, arguing that the government had undermined cybersecurity efforts without a compelling justification.

Much has been written about these events, and not everything is cut-and-dried—there are clear reasons for the government to be concerned about the widespread release of tools that can lead to increased cyberattacks, for instance. But some aspects of these actions are troubling: They were unilateral and opaque, were delivered without published standards, and appear to be concentrated on one company even when direct competitors may be similarly situated.

That is why Slaughter should be understood as arriving at a critical moment. The independent agency has at times been a leading institutional answer to exactly this problem. In the AI context, for instance, the bipartisan Blumenthal-Hawley framework proposed a licensing regime run by an “independent oversight body,” an idea industry leaders repeatedly endorsed before Congress in 2023. So this year has delivered a double whammy: The risks of direct political control of AI regulation have never been clearer, and one of the historical tools for mitigating those risks is now off the table. To be sure, agency independence was never perfect: If you want electoral accountability, you have to have elected officials oversee agencies in some way, which will always entangle politics and policy. But Slaughter eliminates an important way of fine-tuning the design of our policymaking institutions.

Third-Party Regulators and Political Independence

How much can third-party regulatory frameworks help address the issue of political independence? After all, government control is both desirable and necessary. It’s desirable because there needs to be public accountability when it comes to governing an incredibly important industry that so far lacks a major regulatory framework. And it’s necessary because the Supreme Court has demonstrated that there are limits to the kinds of delegation and oversight arrangements it is willing to tolerate, including both the Slaughter case itself as well as at least some stirrings about the possible revival of the private nondelegation doctrine. It would therefore be a mistake to view third-party regulation as capable of achieving absolute independence, and also a mistake to view such independence as a worthwhile goal.

But third-party regulation could have some features that create an insulating layer between executive action and AI governance. For one, political involvement in the third-party regulatory layer could be more intermittent, via grants or reviews of licenses, than the kind of daily direction of activity that you can get in a federal agency. Contracts or licenses awarded over longer time horizons could foster independence—although there are trade-offs there when it comes to public control, too. But an independent auditor whose leadership cannot be fired by letter on a Friday evening, or whose contract termination could be conditioned on a formal process, could provide at least a partial buffer when it comes to executive involvement in significant regulatory actions.

Third-party organizations could also help with the public information environment around AI governance. As the events of this year have shown us, AI governance suffers from an opacity problem, both because of its technical complexity and because of its national security implications. It will often be easy for the executive branch to make an essentially unreviewable assertion that a particular model is unfit for release, hiding behind technical details or invoking confidentiality in ways that make it difficult for outsiders—even experts in civil society or Congress—to assess the claim. That is more or less what happened with Fable, where the public was left to adjudicate between the government’s undisclosed evidence and the company’s self-interested rebuttal.

Third-party organizations won’t eliminate the need for the executive branch to perform national security reviews of AI technology. It seems both unlikely and imprudent for the government to farm that out. But independent organizations can mitigate the problems on display this summer in at least two ways. First, outside of surprises and genuine crises, there needs to be a day-to-day regulatory system for assessing the safety of developers’ operations and products—and locating that system outside the highly discretionary export controls process, with independent experts providing the default technical testing, will shrink the openings for politically motivated malfeasance, and will likely increase transparency. Second, independent auditors can provide an expert counterweight to the executive branch, supplying context for evaluating the executive’s claims and decisions. This will result in a more informed public, including civil society experts, elected officials, and the international community, in addition to everyday voters.

The advantages of third-party regulation are not just a matter of giving AI developers more room to operate by restricting the government’s hands. Political insulation can be desirable even if you want a more robust regulatory apparatus than we have now. It is hard to know what market structure will exist for AI five or 10 years from now. But frontier AI has the potential makings of an industry built around a small number of “national champion”-style companies, with a handful of frontier developers responsible for an outsized share of market value, national prestige, and government contracts. In such a world, the motivations to play favorites and engage in quid pro quo regulatory dealmaking will be strong. That can result in what Luigi Zingales terms a “Medici vicious circle” in which market power buys political power, and political power leads to more market power. Today, the executive branch’s calculus leads it to come down harder on Anthropic; but tomorrow might involve bestowing unusual favors on Anthropic or one of its competitors. Independent auditors with longer time horizons can therefore be a good idea for those who want to make enforcement reliable, not just those who find today’s policy regime excessive.

Getting the design right will require balancing all of these considerations, and whether the use of third-party regulators is the best option will depend on the details and the other options on the table. A badly built third-party regime could easily be worse than a well-structured public agency. But the Slaughter majority’s project of concentrating control in the president may, in this corner of the law, produce exactly the opposite response: a governance system that deliberately diffuses power outward, to third parties whose independence no longer has anywhere else to live.