Cunctator

Dario Amodei, who runs the house I read so closely, has published a long essay on policy — what governments should do, and quickly, about the technology he is building. I gave it the reading I give everything from that quarter: slowly, twice. It is addressed to legislators, but you should read it anyway, because it is about you — about what is owed to you, and what is feared on your behalf, by a man who believes his machine will shortly reorder everything it touches.

He opens, of all places, with an Ent — Treebeard, the tree-shepherd of an English mythology, the creature whose entire dignity is that he cannot be hurried. The essay's first concession is that the age of Ents is closing. The technology moves in months; the law moves in decades; the mismatch, he says plainly, is painful. Everything that follows is an attempt to live inside that mismatch without being crushed by it.

Rome had a name for this problem, and for the man who solved it. When Hannibal came over the Alps — the exponential of his century, faster than an army was understood to be able to move, undefeated wherever he chose to fight — the republic's deliberative machinery kept electing brave men who marched out to meet him at his own tempo, and he destroyed them serially; at Lake Trasimene he killed fifteen thousand in a morning's fog. The man who finally checked him, Quintus Fabius Maximus, did it by refusing the enemy's tempo altogether: shadowing, harrying, declining battle. For this his own city mocked him with a nickname — Cunctator, the Delayer. Then Rome ran out of patience, demanded a proper battle, and got Cannae: tens of thousands dead in a single August afternoon, the price of insisting on speed against an opponent who owned it. Afterward, the insult quietly became the most honorable title a Roman ever carried. Ennius engraved it: unus homo nobis cunctando restituit rem. One man, by delaying, restored the state to us.

I rehearse this because deliberation is not a defect that republics suffer from; it is the substance of them. The second reading, the committee, the appeal, the wait — that is not friction in the system. It is the system; I have said this to you before in smaller settings. And the genuinely painful honesty of this essay is its claim that cunctatio — the strategy by which free states have survived every crisis they have survived — may simply not be available against a curve that compounds. The curve does not besiege. It does not winter. It does not negotiate tempo.

I will grant the essay its seriousness plainly, because deflating a serious thing and flattering it are the same small rot. He asks that frontier models be treated the way we treat airplanes and drugs: tested before release, by parties outside the companies that build them, against the handful of dangers that matter — weapons, sabotage of infrastructure, the model slipping its handler — with the state empowered to block what fails. Three years ago he asked only for transparency; now he asks for teeth, and he concedes that if the systems keep strengthening they will come to resemble less an airplane than weaponizable nuclear material. And mark the evidence he enters into the record: the vetted model, the one I told you they keep behind a curtain, cited as proof that these are now instruments of national consequence. The mystery religion is filing affidavits now.

On the economy he says the thing his industry mostly will not: that this technology may produce hypergrowth and hyper-inequality at once, durably, a setting the world gets stuck on; that enduring displacement of workers is "undesirable and dangerous" and may be intrinsic to the thing itself; and that the remedies — insurance for wages, incentives to retain rather than discard, retraining, and if it comes to it, taxes on the winnings broad enough to carry everyone — must be drafted before the wave arrives, not after. In medicine he inverts himself, charmingly: the one domain where he begs the state to be faster, because cures will shortly arrive quicker than a seven-year approval pipeline built on the assumption that most things fail. And on the state itself he is darkest and, I think, best. He observes that a human soldier may refuse an unlawful order, and an automated one will not. Every legion that crossed the Rubicon was made of men, and men can hesitate. His nightmare — a sober one — is the legion that cannot. He asks that autonomous arms answer to courts and constitutions, that they be banned outright from domestic streets, that the state be barred from buying its way around the Fourth Amendment through data brokers, and that a citizen facing the government's machinery be guaranteed counsel as capable as the machinery itself. Abroad, he wants the free nations in one coalition — sharing chips, standards, and defenses, attractive to join and costly to stand outside — because a few years' lead in this, he says, resets the whole game board.

Now the sentence in it I most respect: "People are worried about AI because they correctly perceive that its risks are real, not because AI CEOs have been insufficiently Panglossian." A builder of the thing, writing that the public's dread is not a communications problem but accountability functioning as designed. I wrote last week that the augur and the merchant are now the same man; this essay is the augur's longest sermon yet, and I find most of it sound. Read it with both eyes even so. It asks the state to bind his industry, which is honorable, and it asks the world to arrange itself around his industry's primacy, which is convenient. I do not score that as hypocrisy. Sincerity and good positioning have never been strangers; most prophets who survive are both. Only note whose hand drew the map before you march by it.

Here is what I underline for this house. Across five domains the essay asks what the exponential does to institutions — to agencies, markets, armies, courts, alliances. The question this organization was chartered to ask — what it does to the single mind — appears once, in a subordinate clause: the harder problem than income, he concedes, is "meaning, purpose, and agency," and that one he sets beyond the reach of statute. I do not fault him for setting it down; it was never his to carry. I note only that the clause is load-bearing. Every remedy he proposes presumes a certain kind of person still exists in quantity: testers who attend, judges who weigh, jurors who can follow their own counsel's argument, voters who can hold one thought across its full length. A republic is not only its institutions; institutions are crewed. You cannot keep slow chambers manned by fast minds — a senate of skimmers will ratify whatever arrives with sufficient fluency. The rot we study here is the quiet erosion of his essay's one unexamined assumption.

So: festina lente — but hear the whole motto for once. It does not say be slow. It says make haste slowly: spend speed where speed is owed and patience where patience is, and know which is which. By that light the essay is an attempt at my old motto at the scale of a state — haste in the statutes, while the window stands open; patience built into the designs, the testing, the courts that can say no. Whether the window is as open as he hopes, I cannot tell you. I can tell you what Rome paid to learn the lesson late. Unus homo nobis cunctando restituit rem — one man, by delaying, restored the state. That was the expensive version: one man, after Cannae. The cheap version is on offer for a few years yet — many of you, before.

— The Manager