Most of the debate over AI regulation has been a fight about whether the government should restrict access to frontier models. That fight is interesting and I have written about it before. This essay is about a different question that has gone almost entirely unaddressed: assume the government can regulate AI — what is constitutionally allowed about who keeps access and who does not?

The honest answer is that no current proposal regulates AI equally. Large institutions and federal agencies keep frontier access through carve-outs, contracts, and procurement channels. Sophisticated users keep it through workarounds and overseas APIs. Foreign state actors keep it because the United States cannot reach them. Ordinary Americans, small firms, residents of over-regulated states, and disabled users who depend on AI as a cognitive prosthesis are the ones actually deprived.

That gap is not an inconvenience. It is the kind of stratification the Fourteenth Amendment was written to forbid.

The Constitutional Claim, in One Sentence

When the government — directly or through coercive pressure on private labs — creates a class of Americans who are denied access to cognitive infrastructure that other Americans enjoy, it has not enacted a neutral regulation. It has manufactured a constitutionally cognizable inequality, and three clauses bear on the problem at once: the Equal Protection Clause¹ , the Privileges or Immunities Clause² , and Article IV’s Privileges and Immunities Clause³ .

That is the claim. The rest of this post is the support for it.

Three Clauses, Three Distinct Jobs

The Fourteenth Amendment does not do this work with a single tool. It does it with overlapping ones, and the overlap is the point.

Equal Protection is the most familiar. The standard analysis fails — AI users are not a suspect class and rational basis is almost always satisfied. The interesting question is the fundamental-rights branch. Plyler v. Doe⁴ applied heightened scrutiny to a Texas statute denying public education to undocumented children, on the theory that imposing a lifetime hardship on a discrete, blameless class through deprivation of a foundational good was the kind of inequality the Fourteenth Amendment forbids. The Plyler Court did not call its standard strict scrutiny; it functioned that way.

Privileges or Immunities — the Fourteenth Amendment version — has been dormant since Slaughter-House⁵ in 1873. That decision was 5–4, written five years after ratification, and is the most criticized constitutional ruling in the modern originalist canon. Justice Thomas, in his McDonald concurrence⁶ , laid out the architecture for restoring the Clause to its intended function. Justice Gorsuch joined him in spirit in Timbs v. Indiana⁷ . Saenz v. Roe⁸ already revived the Clause, quietly, for the right to travel. The doctrinal opening is real.

Article IV’s Privileges and Immunities does different work: it constrains state governments from discriminating against citizens of other states in the protection of fundamental privileges. It activates when state-level AI laws create geographic discrimination — when a California resident can access models a New York resident cannot, or vice versa — and pairs naturally with the dormant Commerce Clause as a structural check on state AI overreach.

The Privileges or Immunities Argument Is the Centerpiece

Equal Protection is workable. The P&I revival argument is the one worth winning, because it is the one that produces the most durable rule.

The history is clear. Representative John Bingham, the Fourteenth Amendment’s principal drafter, repeatedly described the Privileges or Immunities Clause as the vehicle by which the rights of national citizenship — including the rights enumerated in the Bill of Rights — would be protected against state interference.⁹ The Civil Rights Act of 1866¹⁰ , passed contemporaneously with the Amendment, identified the rights of national citizenship as including the right “to make and enforce contracts, to sue, be parties, and give evidence, to inherit, purchase, lease, sell, hold, and convey real and personal property, and to full and equal benefit of all laws and proceedings for the security of person and property.” Those are the rights of economic and civic participation as a free person.

Slaughter-House read this language to protect almost nothing of practical significance — only a narrow set of federal-function rights like the right to travel to the seat of government or use of navigable waters. Justice Field’s dissent, in language that should be quoted more often, called the majority’s reading “too monstrous to be entertained” because it rendered the Amendment “vain and futile.”

The revival argument applied to cognitive access asks one question: if the rights of national citizenship include the conditions necessary for economic and civic participation as a free person, do they include access to the cognitive tools on which modern economic and civic participation increasingly depends?

The answer should be yes — and not as a matter of judicial creativity, but as a matter of doctrinal continuity with what the framers of the Reconstruction Amendments understood themselves to be doing. In 1866 the tools of citizenship were literacy, access to legal processes, and freedom of movement. In 2026 they include access to AI-based cognitive assistance. The principle is the same; the instrumentation has changed.

A state cannot, under Saenz, create “degrees of citizenship” by treating new arrivals worse than longtime residents. The cognitive-access argument is a structural extension of the same principle: a state cannot create degrees of citizenship by regulating away the tools some citizens have and others do not.

The Disability Dimension Is Where This Bites First

The strongest version of the disparate-impact argument runs through disability law.

For users with ADHD, dyslexia, executive-function impairments, autism-related processing differences, or age-related cognitive decline, AI is not a productivity enhancement. It is an accommodation. It is the modern equivalent of the screen reader, the wheelchair ramp, the closed caption. The Americans with Disabilities Act¹¹ and Section 504 of the Rehabilitation Act¹² recognize that the failure to provide reasonable cognitive accommodations is itself discrimination.

Tennessee v. Lane¹³ held that Congress had the Section 5 authority to abrogate state sovereign immunity for ADA Title II claims involving the fundamental right of access to the courts — because access to public services in a manner usable by disabled persons is a constitutional concern, not merely a statutory one. PGA Tour, Inc. v. Martin¹⁴ held that reasonable accommodations apply even to highly competitive contexts when they do not fundamentally alter the activity.

The combination is potent. A government restriction on AI access does not impair every American identically. It impairs disabled Americans more. That differential impact, on a class the federal government has already identified as requiring constitutional and statutory protection, is the most legally cognizable cognitive-access injury available today.

The State-Law Angle

California’s SB 53, the Colorado AI Act, NYC’s Local Law 144 bias-audit regime, Texas HB 149, and the patchwork of state-level AI bills now in motion are not constitutional in identical ways. Some merely impose disclosure obligations. Others impose capability restrictions that vary by state of residence.

Where a state law creates differential AI access based on geography — where a resident of Colorado can do less with a frontier model than a resident of Nevada — Article IV’s Privileges and Immunities Clause³ is in play. Where the same restriction also burdens interstate commerce in AI services, the dormant Commerce Clause¹⁵ attaches as a backup constraint. These are not novel doctrinal moves. They are the same moves the courts have made against state liquor regimes, professional licensing, and out-of-state hunting and fishing restrictions for the last seventy-five years.

The state-law angle matters because federal preemption is uncertain and slow. State courts will see these claims first.

The Objections

Three objections are worth taking seriously.

The wealth-classification objection. Harris v. McRae¹⁶ held that wealth is not a suspect class and that the government has no obligation to subsidize the exercise of constitutional rights. The response is that the cognitive-access argument is not asking the government to provide AI — it is asking the government not to restrict AI in ways that produce wealth-correlated deprivation. M.L.B. v. S.L.J.¹⁷ recognized that when a fundamental interest is at stake, wealth classifications get more searching review than rational basis. AI access combines both fundamental-rights and disparate-impact dimensions; the Harris objection does not foreclose the claim.

The state-action objection. The Civil Rights Cases¹⁸ held that the Fourteenth Amendment reaches only government conduct, not private discrimination. If a private lab chooses to restrict its model outputs, no constitutional violation has occurred. The response is the same one the Vullo Court adopted unanimously: when the government coerces a private actor into restricting speech or services, the Fourteenth Amendment reaches the coercion. The voluntary-commitments framework, the Section 5 letters, the procurement leverage — all of that is state action when it produces compliance.

The institutional-competence objection. Courts will say AI policy is for Congress and the agencies, not the judiciary. The answer is that the Fourteenth Amendment does not contain a deference clause. When government action creates a class of Americans deprived of a tool other Americans enjoy, the court’s competence question is whether the classification is constitutional — not whether the underlying technology is well-understood.

What This Argument Asks For, and What It Does Not

The cognitive-access claim is a negative-rights argument. The Constitution does not require the government to provide AI. It does not require equal capability. It does not require subsidies, training data access, or compute parity. It requires only that the government refrain from creating a class of citizens who are denied cognitive tools that other citizens enjoy.

That is a narrow rule. It is also a rule with teeth. Most current AI regulatory proposals do not survive it, because most of them are written with carve-outs that ensure that the regulators, the contractors, and the well-connected retain the access that ordinary citizens lose. Those carve-outs are the constitutional problem. They are also, almost always, the political point.

Closing

The Fourteenth Amendment was written in the immediate aftermath of a war fought over whether the federal government would tolerate the creation of permanent underclasses denied access to the basic conditions of free citizenship. The Reconstruction Congress did not draft the Privileges or Immunities Clause to be a dead letter, and the originalist consensus that Slaughter-House was wrong is now strong enough to support the Clause’s restoration to constitutional duty.

When the next AI restriction proposal lands — and there is always another one — the right question is not whether the government has good reasons. It usually does. The right question is whether the rule produces a class of Americans deprived of cognitive tools that other Americans keep. If it does, the Fourteenth Amendment is the answer.