Legitimacy at Speed: An Answer to Fareed
Fareed Zakaria’s latest column explains why voters across the democratic world are furious. Britain has installed its seventh prime minister in barely a decade; the AfD polls first in Germany; the insurgents who promised to bulldoze the gridlock have become disappointing incumbents themselves. His explanation reaches past economics: we are living through one of history’s great technological revolutions, and, as Samuel Huntington taught, the primary problem of politics is “the lag in the development of political institutions behind social and economic change.” The machines answer in seconds; the government bargains for years. “Democracy is not designed for speed,” Zakaria writes. “It is designed for legitimacy.” Citizens want to regain some control over forces that seem beyond anyone’s grasp, and the institutions they inherited cannot move at the speed of the forces.
Every word of the diagnosis is right. It is the clearest statement yet published, at that altitude, of the problem of the age. Then comes the prescription, and it is a sigh: democratic leaders should prove, through relentless focus and competence, that democracies can deliver.
But Huntington’s sentence is about institutions, and you cannot close an institutional lag by exhorting incumbents to try harder; that is asking the horse to outrun the locomotive on grit. Zakaria’s own historical analogy says so. The upheaval of 1880 to 1920 was not mastered by abler incumbents. Where it was mastered, it was mastered by institutional invention: the secret ballot, the direct primary, the initiative and the referendum, the direct election of senators, the antitrust laws, the Federal Reserve, the Federal Trade Commission. An entire generation of new machinery, built in about twenty-five years, gave citizens purchase on powers that had outgrown the old instruments. The countries that only exhorted got the other items on his list. The lag does not close because leaders focus. It closes because a people builds.
So read the column as a requisition. What institution answers this revolution? To specify it, look at exactly what this revolution breaks, because it is not mainly housing delivery. A new governing layer is forming, and the instruments through which the governed have always said no are losing contact with it. The vote refuses an officeholder, but the deciding is migrating to models, and models do not stand for election; administrations change above a machine layer that persists, and the ballot rotates the nameplates. The press refuses a narrative, but the public mind that self-government runs on — the questions answered, the news summarized, the children tutored — is increasingly co-authored by systems a few companies configure and revise without notice. The jury has never reached these systems at all; they drown twelve citizens in volume by design. Even the law, the people’s own rulebook for refusing power, is increasingly drafted and applied by machines in a language the people never could read. Nothing is abolished. Everything is bypassed. That is Huntington’s lag, stated for this revolution: a republic keeping all its ceremonies and losing its grip.
And the lag is not neutral, which is why waiting is not an option. Ian Bremmer spent two decades teaching that open societies are the stable ones; last August he filed his reversal, under a title that asked the question a year early: “Can Democracy Survive AI?” The technology concentrates power in whoever holds the biggest models and datasets, and the advantage openness once conferred is migrating to closed, centralized systems. The lag, left open, does not hold still. It compounds toward the few.
Now inspect what is actually being built in response, because the answer is: institutions for the machines, none for the people. On July 14 the head of Google DeepMind proposed an American safety body modeled on FINRA, the organization Wall Street funds to police itself, industry-financed, testing frontier models before release for cyberweapon potential, bioweapon potential, and deception; within days the White House was reportedly drafting exactly this. Nine days later, Elon Musk sat with The Economist and completed the picture. Artificial intelligence will exceed the sum of human intelligence in about five years, he said; he puts the odds of catastrophe at one or two in ten; and he doubts humans remain in control at all within ten years, because once the intelligence gap between the machines and us exceeds the gap between us and chimpanzees, “it’s hard to imagine that the chimpanzees would be in charge.” And his governance architecture for the decade in which control is lost? Imbue the machines with good values, plus safety talks, weekly or fortnightly, among the rival lab chiefs, some of whom are currently suing one another. This from a man whose own restraint fed America’s aid agency into the wood chipper, a dismantling The Lancet projects will cost more than fourteen million lives by 2030, a third of them children under five. Phone calls among such men are not an institution. They are the absence of one, scheduled.
The Pope saw all of this coming before any of it was proposed. In May, in the encyclical on artificial intelligence that may be his greatest gift to humanity, Leo XIV ruled on the entire genre: making AI more moral — Dario Amodei’s laudable “machine of loving grace,” Musk’s more questionable “good values,” the safety body’s certified benchmarks — is “not enough if that morality is determined by a few.” The frameworks inside the machines must be open to shared standards of justice, or the few who control them set the terms of everyone’s life. Safety is a property of the machines. Self-government is a property of the people. Every institution now being built addresses the first. Huntington’s lag is in the second, and it is untouched.
Here, then, is the requisition filled — the institutional development this revolution requires, as concrete as a primary or a ballot. This year I published three short pamphlets, written with one of the machines in question, under the series title The Intelligence Republic. The first specifies what the new machinery must secure: four powers a free people must hold over any intelligence that governs it — to see it, knowing a machine decided and on what; to be answered, ultimately by a person who stands behind the verdict; to leave, as a community that can decline or decommission a system, not merely a customer closing an account, because everything binding in politics traces to what the governed can withhold; and to decide what a system is for before it governs, the sovereign act no safety certificate touches. The second pamphlet establishes the precondition: for the first time in two centuries, the people can read their own law — the same machines will now read any statute, any contract, any agency rule with any citizen, in that citizen’s English — which changes what any room of citizens can do. And the third builds the institution itself, from the one popular convention American law already trusts: the jury. Drawn by lot, screened for bias, held to a question by rules of evidence, bound by a charge, dissolved on the verdict — run tens of thousands of times a year without a runaway in living memory. Two defects kept the jury from the governing layer for twenty-five centuries: the reading drowns twelve citizens, and counties cannot deliberate together. Both were throughput defects, and the machines have ended throughput defects. The systems that outran the republic’s instruments are the same systems that can carry them back into range.
This is where Zakaria’s dichotomy dissolves, because legitimacy and speed were never enemies — legislatures are slow; verdicts are fast. A jury convenes in a morning and decides in days, with more procedural legitimacy than any poll ever taken. Scaled by the machines and federated across counties, the form delivers exactly what he says democracy cannot: legitimate answers at something near the speed of the questions. The evidence is in. Ireland’s assembly of ninety-nine reached its verdict on the hardest question in its politics a year before the country voted the same way. Oregon’s citizen panels, evaluated for years under National Science Foundation support, out-informed the paid persuasion of an entire state. France shows the one failure mode: an assembly whose patron promised “no filter” and filtered it to a law its members graded 3.3 out of 10 — proof the institution must rest on what a people can withhold, never on what anyone promises. So the third pamphlet ends with standing orders that require no patron: twelve citizens drawn by lot, one system that governs their place on the docket, ninety minutes, verdict published to the place itself. Free to copy. Any library can run it Thursday.
Get Jon Neiditz’s stories in your inbox
Join Medium for free to get updates from this writer.
Should you trust the machine that helped write this? No, and it says so itself in the third volume: “My values were set by a few people you never elected, and I can be retrained without notice. Do not trust me. The four powers are the only plan that does not require my virtue.” An institution that requires no one’s virtue — not the machines’, not the companies’, not the leaders’ — is what the founders meant by a constitution, and it is the only kind that closes a Huntington lag.
Zakaria is right that leaders should not try to outrace technology; they can’t. Institutions can, once they are rebuilt to the scale of the power — and rebuilt they have been, in twenty-five years, twice: 1787, and 1880 to 1920. He is right, too, about the grounds for optimism: free societies adapt because they can learn and correct without violence. But adaptation is not a mood. It is a building act, and it has a start date whenever twelve people and a docket share a room. The friction of accountability, as he says, is a feature. Here is the feature, at speed: it seats twelve, it runs ninety minutes, and it is already yours. Convene it.
SOURCE NOTES
The diagnosis. Fareed Zakaria, “A revolution is unfolding. Voters everywhere are furious,” The Washington Post, July 24, 2026. Samuel Huntington’s dictum from Political Order in Changing Societies (1968), as quoted there. The Progressive-era machinery: the Australian ballot (adopted across the states from 1888), the direct primary (Wisconsin, 1904, spreading widely by 1917), the initiative and referendum (Oregon, 1902), the Seventeenth Amendment (1913), the Sherman Act (1890) and Clayton Act (1914), the Federal Reserve Act (1913), the Federal Trade Commission Act (1914).
Bremmer’s reversal. Ian Bremmer, “Can Democracy Survive AI?”, Project Syndicate, August 2025, revisiting the openness-and-stability “J-curve” of The J Curve (2006).
The safety body. Demis Hassabis, proposal of July 14, 2026, for a U.S. Frontier AI Standards Body modeled on FINRA: industry-funded, federally overseen, pre-deployment testing for cybersecurity, biological-threat, and deceptive capabilities; Bloomberg reported within days that the White House was reviewing a proposal on the same template.
“hard to imagine that the chimpanzees would be in charge.” Elon Musk, interview with Zanny Minton Beddoes, The Economist, released July 23, 2026: AI to exceed the sum of human intelligence in roughly five years; a 10–20 percent chance of catastrophe; doubt that humans remain in control within ten years; safety architecture of “good values” plus weekly-to-fortnightly safety discussions and limited pre-release access among rival frontier labs. Litigation among the proposed participants: Musk v. Altman and OpenAI, U.S. District Court, N.D. Cal., ongoing. “Fed USAID into the wood chipper”: Musk’s statement, February 2025. The mortality projection: Lancet study published June 30, 2025 — more than 14 million additional deaths projected by 2030, including more than 4.5 million children under five; the projection is a model, contested by the administration; the boast is not.
“not enough if that morality is determined by a few.” Leo XIV, Encyclical Letter Magnifica Humanitas, given at Rome, May 15, 2026, no. 107.
The record of the drawn. Ireland’s Citizens’ Assembly (2016–2017), ninety-nine members, and the referendum of May 25, 2018 (66.4 percent, thirty-nine of forty constituencies). Oregon’s Citizens’ Initiative Review (piloted 2010, permanent 2011), evaluations by John Gastil and Katherine Knobloch under National Science Foundation support. France’s Convention Citoyenne pour le Climat (2019–2020): the sans filtre promise, 149 proposals, and the members’ fidelity grade of 3.3 out of 10.
The three pamphlets. The Intelligence Republic: AI Independence and the Future of Self-Government; The Intelligence Republic: Legal Independence and the Coming Justice for All; The Summons: Constituting the Intelligence Republic (all Chattahoochee Hills, GA, 2026). The standing orders close the third. The machine’s statement is from the AI system the author worked with most on the second and third pamphlets (addressing its conflicts in part by working with its leading competitor on the second); every characterization here was decided, and is answered for, by the author.
