— A Review of From Traditional Regulation to Verifiable Regulation
From Traditional Regulation to Verifiable Regulation appears to be about how the U.S. Securities and Exchange Commission should respond to AI and crypto. At a deeper level, however, it asks a larger question: under what technological conditions did the regulatory system we now take for granted arise? And if those conditions change, should regulation change with them?
The article does not begin with the familiar question of whether regulation should be stricter or looser. That debate has gone on for years, usually with both sides assuming that the institutional form of regulation is fixed and that only the amount of regulation is in dispute.
The article goes one level deeper. Why must banks be audited? Why do securities firms need licenses? Why are trading, custody, and clearing separated? Fraud prevention and investor protection are obvious answers. But there is another practical reason: historically, market participants could not independently verify many of the facts on which financial trust depended.
Depositors could not inspect a bank’s assets for themselves. Investors could not verify every line of a financial statement. Customers could not see what was happening inside an exchange. Regulators themselves could not watch every transaction continuously. Society therefore built organizations to certify facts and then built further organizations to monitor those organizations.
The article compresses this history into one powerful proposition: regulation uses organizational cost to compensate for verification cost.
The strength of this formulation is that it does not dismiss traditional regulation. It explains why traditional regulation made sense. Many important facts were difficult to see, expensive to check, and impossible to verify continuously. Digital records, cryptography, real-time interfaces, and AI are now lowering those verification costs. If some institutions were created partly to cope with high verification costs, then some of those institutional arrangements deserve to be re-examined as the cost structure changes.
That goes much further than saying that AI can make regulation more efficient. Efficiency means putting a faster machine inside the old system. The article asks a more difficult question: if the conditions of verification have changed, do some parts of the old structure still need to remain exactly as they are?
Bitcoin therefore takes on a meaning different from the usual shorthand of “decentralization.” It did not eliminate power. Developers can propose new versions, miners and nodes can choose whether to adopt them, and users can make their own choices. What it demonstrated, however, is that checks on power do not always have to take the form of one institution supervising another. Public rules, choice, and exit can also constrain behavior.
This moves verifiability from a technical issue into an institutional one. If a fact can already be verified directly by participants, do several layers of institutions still need to certify the same fact? If market participants can act under public rules and exercise meaningful choice, can some forms of constraint that once required administrative organization be shifted partly toward rules and market mechanisms? The article does not answer yes automatically. It treats this as a question worth testing, which is the right degree of restraint.
A second important distinction is between transparency and verifiability. Transparency asks whether something can be seen. Verifiability asks whether it can be independently checked. An institution may publish hundreds of pages of disclosure and still leave the reader dependent on the institution’s own word for the most important numbers.
This is not merely a difference in vocabulary. It changes the way credit can be formed. In the past, trust often depended heavily on who was speaking: a government, a bank, an auditor, a rating agency. Verifiability adds another possibility. I do not trust you only because you possess a license or a reputation; I gain an additional reason to trust you because the critical facts can be independently confirmed.
A third important move is to treat regulatory rules themselves as judgments. When regulators face a technology with little historical precedent, they cannot know all future facts when the first rules are written. Rules therefore contain assumptions, and assumptions can fail.
The article responds with the idea of Minimum Viable Regulation: define the objective, boundaries, and failure conditions; run a limited real-world experiment; observe the result; then expand or revise. The importance of this idea is not its borrowing from software vocabulary. It is the willingness to state openly that regulators can make judgment errors.
Error is not the greatest danger. A system that cannot discover its own error is.
That is why regulatory rules should also have versions. Why was a rule adopted at the time? What facts were known? What outcome was expected? Why was it later changed? Those records should remain available. Flexible regulation can easily become one interpretation today and another tomorrow. Versioned regulation is different: it may change, but it cannot erase its own past.
The article then takes another step: the verification system itself must be verifiable.
Any theory of verification immediately faces the question: who verifies the verifier? If the answer is simply to build a new verification center and ask everyone to trust it, the problem has merely been recreated. The article explicitly recognizes that verification infrastructure will itself become a new locus of power. Who defines the interface? Who controls the underlying data? Who certifies machine-readable rules? Who bears responsibility when verification fails? The sources, permissions, versions, and responsibilities behind those systems must themselves be reviewable.
A theory closes the loop only when it is willing to subject its own instruments to the same standard it demands of others. Otherwise, “verifiability” becomes a slogan imposed outward.
If the claim that regulators must also be verifiable is translated into a minimum institutional design, at least four elements follow. Important rules should retain complete version histories. Major regulatory judgments should preserve the evidence and reasons available at the time, including reasons for later revision. Official AI outputs should distinguish informational responses, advisory opinions, and decisions on which regulated parties may reasonably rely, with clear routes for human review and appeal. And the interfaces, data sources, and machine rules used for verification should themselves be subject to independent audit.
This is also where the hardest questions begin. Who has authority to define verification standards? Who oversees the verification infrastructure? How should courts review machine-executable rules? Which disclosure obligations can shrink when facts become continuously verifiable, and which obligations should become stricter? How should old and new systems coexist? Technology does not answer these questions automatically. They concern authorization, procedure, responsibility, and the redistribution of institutional power. That does not weaken the article. It shows that the discussion has moved from regulatory philosophy to institutional interfaces.
The article is similarly careful about AI. It does not claim that AI should decide what counts as a security, nor that machines should replace regulators. Facts can be verified where possible. Processes can be replayed. Judgments involving legal interpretation, competing interests, and value choices remain human responsibilities. AI’s role is not to abolish judgment, but to clarify as much of the factual ground as possible before judgment begins.
The article therefore does not deliver a regulatory bill ready for a congressional vote. Its achievement is to change the starting point from which regulation is examined.
We are accustomed to asking: what rule should be written?
The article adds several questions that come before that one. Can the underlying facts now be verified? If they can, how much of the old institutional cost is still necessary? If they cannot, how should power remain divided? After a rule is implemented, does anyone return to test the assumptions behind it? Can the regulator’s own judgments—and the systems used to verify others—be replayed and reviewed?
Traditional regulation has largely said: I have the authority to require you to prove yourself.
Verifiable regulation adds another sentence:
I should also be able to explain where my judgment came from, and future facts should be allowed to come back and test it.
If that principle ever enters real institutions, its significance will extend far beyond how the SEC regulates crypto.
