Who Owns the Gate

As a rule, regulation is acquired by the industry and is designed and operated primarily for its benefit.

Once authority becomes scarce, the right to supply it becomes an economic position. A professional license, operating permit, certification, franchise, insurance policy, or access agreement may perform indispensable public work. It may also become a toll. Often it is both.

This ambiguity is the real terrain of capture. If diagnostic capability becomes cheap while prescriptions still require a physician, part of the physician's compensation may shift away from diagnosis and toward the authority, judgment, and exposure attached to signing. That payment is not automatically rent. Someone must examine the case, refuse unsafe recommendations, communicate uncertainty, and remain answerable. Rent begins where law preserves the exclusive gate after those purposes could be achieved more safely or openly by another arrangement.

The engineer's stamp, lawyer's opinion, accountant's attestation, and regulator's certification present the same question in different institutional forms. What does the gate actually contribute? Does it test capability, examine this case, identify a principal, create an insurance surface, or simply limit entry? The answer can change as computation changes the work beneath the credential.

Stigler's theory matters here because capture does not require a conspiracy.¹ Concentrated beneficiaries can organize around rules whose costs are distributed among people with little reason to contest any single proceeding. Once a restriction exists, expected rents can be capitalized into credentials, firms, and franchise positions. Reform then appears to confiscate an investment rather than remove a privilege.²

Yet a theory of capture can itself become lazy. It is not enough to observe that an incumbent supports a rule. Incumbents may know risks that entrants prefer to discount. Nor does a measurable performance advantage settle the issue. A system that diagnoses accurately may still fail in communicating uncertainty, monitoring distribution shift, protecting privacy, or responding when a patient does not resemble its evaluation population. Safety is not a ceremonial word, and performance is not one number.

The useful distinction is functional. A rule should be able to say which harm it addresses, why the proposed gate can reduce that harm, what evidence would justify wider entry, and who answers when the gate itself fails. Requirements that cannot survive those questions are candidates for capture even when expressed in the language of care.

The history of automated legal documents shows how a boundary can move. In 1999, the Texas Unauthorized Practice of Law Committee obtained a trial-court ruling against software that supplied legal forms. The appeal became moot after the legislature exempted software that clearly disclaimed substitution for legal advice.³ The episode did not prove that all professional restrictions are protectionism. It revealed that a category written around human practice could be redrawn once software began performing part of the work.

That redrawing never ends. A form generator may supply language without advising a client. A model may select provisions, ask questions, or tailor the instrument to circumstances. At some point the product crosses from transmitting a form into exercising judgment with legal consequence. The difficult task is to locate that point without defining every useful computation as professional practice and without pretending that a disclaimer removes the need for responsibility.

Three broad regulatory strategies recur. An ex-ante regime requires evidence and permission before deployment. An ex-post regime permits entry but assigns liability after harm. A supervised-entry regime widens authority gradually, conditioning it on audit, bonding, insurance, reporting, or limits on use. These are not ideological camps. A domain may need all three at different layers.

Ex-ante review is strongest where harm is difficult to reverse, affected people cannot meaningfully consent, and failure may not become visible until it has spread. Its danger is that the approving office becomes a bottleneck whose criteria reflect incumbent practice and whose delay is borne by outsiders. Ex-post liability encourages experimentation where injury is observable and compensable. Its danger is that a judgment-proof deployer can impose losses it will never pay. Supervised entry can connect freedom to evidence, but only if the conditions are intelligible and the path to broader authority is real.

Time cannot settle the choice by itself. Delay may prevent the adoption of a dangerous system, or prolong exposure to a worse one already in use. The comparison must be between reachable states, not between innovation and an imaginary condition of zero risk. A precautionary regime that never measures the harm of its own delay is not precautionary enough.

This is where capture becomes hard to distinguish from governance. An incumbent can defend a valuable safeguard because it protects the public and because it protects the incumbent. A deployer can attack a useless restriction because it blocks innovation and because it blocks the deployer's revenue. Institutions should be designed on the assumption that mixed motives are normal. Transparent standards, independent evidence, reviewable decisions, time limits, and credible routes of appeal matter more than claims to purity.

Jurisdiction as an Input

Computational services can cross borders more easily than hospitals, power plants, and courts. This difference creates room for jurisdictional arbitrage, though much less than slogans about borderless software imply. A company may incorporate in a permissive jurisdiction while still depending on local data, licensed professionals, banking access, physical infrastructure, and the courts of the place where harm occurs. Law travels through those dependencies.

Regimes nevertheless compete. A jurisdiction can attract deployment by making rules legible, approvals timely, and liability insurable. It can also become a haven by externalizing harms onto people elsewhere. Restrictive states may answer with data localization, access controls, local-agent requirements, or liability for domestic distribution. Whether arbitrage disciplines a bad gate or evades a good one depends on what the gate was doing.

The direction of convergence is therefore open. Permissive regimes may demonstrate that incumbents exaggerated risk. A visible failure may instead harden restrictions across borders. Large markets can export standards because access to their consumers is valuable. Small states can influence architecture by offering legal homes to firms that would otherwise lack one. The result is a contest among institutional packages, not a frictionless race to whichever jurisdiction asks least.

This contest changes technology itself. Developers optimize for the evidence regulators require, the harms insurers price, and the uses licenses permit. An approval regime does more than slow or accelerate a fixed trajectory. It helps choose which systems are built, what they disclose, and whose risks count during design.

The political economy is asymmetrical. A gate creates an office, and an office creates people with reason to preserve it. Distributed beneficiaries of wider entry may never organize with equal intensity. But abolition also creates constituencies, and concentrated deployers can capture permissive regimes as readily as incumbents capture restrictive ones. The conflict is not regulation against markets. It is among rival arrangements of permission, proof, and loss.

The right question is not whether a human remains in the loop. It is whether the remaining institution can still see the act, refuse it, answer for it, and repair its consequences. Where those powers exist, a gate may deserve its authority. Where they have been removed while the toll remains, the gate has become property masquerading as judgment.

Who owns that property will help decide where the surplus goes. Who governs it will decide whether cognitive abundance widens the field of action or merely makes admission to the field more valuable.