The visa the model refused

Consider an applicant who files for a visa from abroad — papers in order, invitation attached, fee paid — and receives, weeks later, a single line: the application is refused, and no further explanation will be provided. Or the version that leaves less trace: the traveler arrives, hands over a passport, and is quietly routed to a room for secondary screening while the line moves on without them. In both cases a determination has been made, and in both cases it rests, somewhere in the machinery, on a risk score the person will never see, assigned by a system they will never be told about, applied under an authority that courts have long agreed to review with a very light touch. They can appeal to no one who can tell them what the score said, because no one who could tell them is obliged to — and, increasingly, no one who could tell them entirely knows.

This is a composite, and deliberately generic — no country, no agency, no system, no person. It has to be, because the point is not any single border but the structure that every border shares. The power to decide who may enter is old, it is broad, and it has always been exercised with less external scrutiny than almost any other decision a government makes. When we bolt an unseeable model onto that power, we are not adding automation to an ordinary administrative act. We are adding opacity to a decision that was already among the least accountable a state ever renders.

The oldest discretion meets the newest opacity

Start with the discretion. The authority to admit or exclude a non-citizen sits close to the definitional core of sovereignty — a state that could not control its own frontier would struggle to be called a state at all. Courts across many traditions have treated this power as one they should touch lightly, deferring to the political branches on questions of who belongs inside and who may be kept out. Whatever one thinks of that deference, it is a settled fact of how border decisions are reviewed, which means the ordinary machinery of contestation — the demand for reasons, the hearing, the appeal on the merits — runs thin here by design. The frontier is where a government is most sovereign and least second-guessed.

Now add the applicant's position, which is the mirror image of the state's. The person at the border is very often a non-citizen, physically abroad or not yet admitted, holding few of the procedural rights that attach to people already inside. They may have no right to counsel at the decisive moment, no right to see the evidence, no forum in which to compel an answer. Their weakness is not incidental to the situation; it is part of what makes the border the border. So even before any model enters the story, we have a decision made under broad discretion, shielded from ordinary review, against a party with almost no standing to contest it.

Automation adds the third strain, and it is the one that changes the character of the whole arrangement. A risk score generated by a model is not merely another factor in the file; it is a factor no one in the room can fully account for, applied to a person who could not contest it even if it were legible. Each of these features — sovereign breadth, weak standing, machine opacity — already strains accountability on its own. Stacked together, they do not add. They compound. The discretion excuses the state from explaining, the applicant's position excuses it from being asked, and the model ensures that even a state inclined to explain may not be able to.

A border is where a state is most powerful and least watched, and a model deployed there inherits both — the power to decide a life and the freedom from having to say why.

Legitimacy where authority is not in doubt

It is important to be clear about what is and is not in question here, because the honest version of this argument concedes a great deal. The state's authority to refuse is not the problem. A government may decline to admit a person, and it may do so for reasons it is entitled to keep to itself, and it may weigh security judgments that no applicant has a right to inspect. None of that is under attack. This essay is not an argument that borders should be open, or that sovereign discretion is illegitimate, or that every refusal owes a full accounting. The distinction it turns on is the one between authorization and legitimacy — between being within your rights to act and being answerable for how you did.

Authority answers the question may you do this. Legitimacy answers a different one: can you be held to account for having done it. A state can possess the first in full and lack the second entirely, and the border is exactly where that gap opens widest. A life-shaping refusal — a family kept apart, a job forfeited, a return home made impossible — decided by a score that no human examined and no procedure will disclose is authority without an account. It is not that the state acted beyond its powers. It is that the state exercised its powers in a way that cannot be answered for by anyone, to anyone, ever. The decision has an author in law and no author in fact.

This is why automating a power already shielded from review is a special danger, and not merely more of the same. The light-touch tradition of border review rested on an implicit assumption: that behind the unreviewable decision stood an official who at least knew why it had been made, even if never compelled to say. Deference presupposed a human account held in reserve. An unseeable model dissolves that reserve. The decision becomes not merely unreviewed but, in the deepest sense, unaccounted-for — there is no held-back reason waiting to be produced under pressure, because the basis was never assembled in a form a person could give. The oldest justification for not asking was that someone, at least, knew the answer. Automate the discretion and even that is gone.

What legitimacy would still require

The reflexive objection is that security forbids all of this — that the enterprise depends on secrecy, and any demand for an account is a demand to disarm. That objection deserves to be taken seriously, because it is sometimes true. There are inputs to a border decision that genuinely cannot be disclosed without compromising sources or methods, and a legitimacy argument that pretends otherwise is not worth making. But secrecy about the basis of a decision and accountability for the fact of it are not the same thing, and the strongest move here is to separate them cleanly.

What legitimacy requires, even where full disclosure is off the table, is a record of the basis sufficient for some independent review to occur — not necessarily by the applicant, but by someone. A determination that a person is adverse to the national interest can remain confidential in its particulars while still generating an auditable record: what the model was, what it weighed, what it returned, and who, if anyone, looked at the result before it hardened into a refusal. That record separates cleanly into parts. There is the material that must stay secret, which can. There is the bare fact of an adverse determination, which the person can be told without unraveling anything. There is a route to human reconsideration, so that no life-shaping decision is closed by a score alone. And there is a preserved trail — a Decision Receipt, in the vocabulary this series uses — that some inspector general, court, or oversight body could later replay, even if the affected person never can.

None of that opens the border, discloses a source, or ties a government's hands. What it refuses to concede is the last combination — opacity plus discretion plus no standing — because that combination is precisely where legitimacy disappears entirely. Take any one of the three away and something is recoverable. Keep the discretion but demand a record, and there is an account, held in reserve, the way the old tradition assumed. Keep the opacity but grant some standing, and there is a party who can press. Keep the weak standing but strip the opacity, and at least a human answered for the call. It is only when all three hold at once that a state can decide a life and owe the decision to no one at all. That is not sovereignty exercised. That is accountability quietly abolished under cover of it.

The scenario above is illustrative — a composite drawn to show a pattern, not an account of any real person, country, agency, system, or event.

— Dispatches · Summit Cognitive