Horizon Accord | Bernie Sanders | Economic Power | Machine Learning
The Uses and Limits of Universalism
Bernie Sanders names economic power more clearly than almost anyone in American politics. This is a study of what that instrument does when it meets an industry that profits from being believed.
IThe Universalist
There is a specific thing Bernie Sanders does better than almost anyone else in American public life. He looks at a situation most politicians describe as complicated, and he names the money. He says the word billionaire where others say stakeholder. He identifies who owns a thing, who profits from it, and who pays for it, and he does this consistently, in the same plain register, whether the room wants to hear it or not. For a generation of Americans who suspected that establishment politics was engineered to keep them from noticing exactly that, this clarity was the whole appeal. It still is.
This essay begins by taking that strength seriously, because the analysis that follows does not work otherwise. If Sanders's framework were empty, there would be nothing here to study. The interest lies precisely in the fact that it is not empty — that it is often more penetrating than what surrounds it. His politics center a small number of structures and return to them with unusual discipline: concentrated wealth, the power of employers over workers, the absence of universal social provision, the machinery by which a few people convert economic advantage into political control. When a harm can be described in those terms, Sanders describes it with force, and he is frequently right.
A framework is not just a set of beliefs. It is a set of instruments — a way of looking that makes some things vivid and lets others fall out of focus. Sanders's instrument is calibrated to detect economic power operating on a universal scale: harms done to everyone, or to everyone who works, by the few who own. That calibration is his gift. What this essay examines is narrower and more interesting than whether he is right about economics. It concerns the instrument itself, and what it does at its edge.
The proposition to be tested is not that Sanders is secretly conservative, or indifferent to marginalized people, or a fraud. Those are smaller and weaker stories, and the evidence does not support them. The proposition is structural: that Sanders's universalism is unusually powerful when a harm can be rendered as a universal class problem, and becomes less reliable when justice requires treating a specifically situated form of domination as primary in its own right, rather than translating it into economic language. That is a claim about an analytical reflex, not a claim about a man's heart. The rest of this essay tries to find out where it holds, where it breaks, and what it predicts.
IIThe Critique From the Left
The sharpest version of this question was not posed by a conservative. It was posed in 2016 by Angela Davis — the Black feminist, prison abolitionist, and lifelong Marxist. Asked about the presidential race on Democracy Now, Davis declined to endorse Sanders or anyone else. She said the issues he raised were "absolutely important." And then she named what she believed his framework could not do: she described a tendency toward economic reductionism, and argued that Sanders lacked the vocabulary to treat racism as something built into the structure of capitalism rather than as an additional problem that economic inequality happens to produce.
It is worth slowing down on the distinction Davis was drawing, because it is easy to flatten into "class versus race" — which is not what she said. Almost everyone agrees that class and race both matter. The real question is about architecture. In one account, capitalism is the engine and racism is a kind of exhaust — an ugly byproduct that a sufficiently powerful economic remedy would reduce along with everything else. In the other account, racial hierarchy is not exhaust but part of the engine: capitalism as it actually developed was organized through racial domination, and a politics that treats race as downstream will keep arriving at solutions that quietly leave the racial structure running. Davis was locating Sanders in the first account and saying the second is the true one. That is not a complaint about his sincerity. It is a claim about the limits of his instrument.
If that were the whole story, the essay would be simple. It is not, and the complication is essential rather than decorative. Some of the most serious Black feminist, socialist, queer, and abolitionist thinkers in the country did not merely tolerate Sanders — they actively supported him. Keeanga-Yamahtta Taylor, the scholar of Black liberation politics, described herself plainly as a Sanders supporter. Barbara Smith, co-founder of the Combahee River Collective and one of the people who helped give intersectional politics its original language, endorsed him in her own name, saying his commitments most closely reflected the commitments she had held her whole life. In 2020 a public feminist statement backing the movement to elect Sanders carried the names of Barbara Smith, Beverly Smith, Barbara Ransby, the abolitionist lawyer Derecka Purnell, and Black Lives Matter co-founder Opal Tometi, among others. Patrisse Cullors, who declined to endorse him in 2016, signed on in 2020.
The record on the other side has to be stated with the same care, because it is easy to conscript. Alicia Garza, as a co-founder of the Black Lives Matter network, stated that the network would not endorse any candidate; that is a documented organizational position and should not be inflated into a personal verdict on Sanders. Mariame Kaba and Ruth Wilson Gilmore, two of the most influential abolitionist thinkers alive, cannot be reliably placed on either side from the public record; the honest entry is that no verified presidential position was found, and an absence of endorsement is not a position. Hillary Clinton was a real candidate in that 2016 race with substantial feminist support of her own, so no one's alignment can simply be inferred from the field. None of these people were witnesses for a prosecution. They were serious analysts making political choices in a real election.
It matters exactly how these two facts fit together, because it is tempting to make them cancel — to treat the endorsements as a refutation of Davis. They are not, and the reason is worth stating plainly. Supporting a candidate and agreeing with an analysis of that candidate are different acts. A person can conclude that Sanders's redistributive program would materially improve Black, poor, queer, and working-class lives — and vote for him on exactly that ground — while also believing his political analysis is economically reductionist in precisely the way Davis described. Davis herself is the proof of concept: she called the issues his campaign raised "absolutely important" and credited it with useful pressure, in the same breath as the critique. So the endorsements do not establish that these thinkers rejected Davis's formulation. What they establish is narrower and more interesting: that the limitation Davis identified was, for many serious people, not politically disqualifying. The value of a program and the resolution of an instrument are two different questions, and this essay is about the second one.
That is the tool this essay will use. Davis supplies the precise formulation — constitutive, not merely produced — and the mixed reception tells us the formulation is compatible with real political support rather than refuted by it. This section only sets the terms. Whatever answer exists has to come from the record of what Sanders actually did when his framework met a harm that would not fully translate.
IIIWhere Universalism Works
Before testing where the framework strains, honesty requires establishing where it succeeds — decisively, and often before it was safe to do so. In 1996, Sanders voted against the Defense of Marriage Act, at a time when the political consensus, including much of his own party's leadership, ran the other way. The same year, he voted against the welfare reform law whose benefit caps fell hardest on poor women and children of color. In 2001 he voted against the USA PATRIOT Act, and against its reauthorization. In 2002 he voted against authorizing the Iraq War, and said on the floor why. These are not small votes, and several carried real political risk at the moment they were cast.
The temptation is to file these as points in Sanders's favor and move on. That is the wrong use of them. They are not character references; they are evidence, and they do specific analytical work: they establish that the proposed pattern cannot simply be "Sanders fails to see marginalized harm." He plainly does see it, and acts on it, sometimes at cost. So whatever limitation this essay is testing has to be more specific than neglect. One distinction is available in the shape these votes share: in each case, the harm to a marginalized group could also be stated as a violation of a universal principle — equal protection, civil liberty, the wrongness of a war, the duty not to abandon the poor. These cases are at minimum compatible with a framework that reads a situated harm clearly when it can be expressed through one of its established universal principles.
This is why the counterevidence narrows the thesis rather than destroying it. The pattern under test is not that Sanders neglects the marginalized. It is more specific: that his framework performs well when a situated harm can be carried by a universal principle — and that the strain, if it appears, will appear where that translation leaves a residue, where centering the specific group is the thing justice requires and no universal restatement quite captures it. Whether that distinction holds is what the next sections test.
IVThe Problem of "Specifically"
In April 2019, Sanders appeared before the She the People forum in Houston, in front of a large audience of women of color. Asked what he would do about the rise of white-supremacist violence, he answered by recalling that he had marched with Dr. King in 1963. The crowd's response was audible and cool. When the moderator Joy Reid pressed him on how he would win the support of Black women, and then pressed again — "and for black women specifically?" — Sanders asked her to repeat the question over the noise, then returned to general language about justice. Afterward, his own senior adviser Nina Turner, a Black woman, said on television that she did not know what had happened on that stage and wished he had answered differently.
This was not the first time. By multiple accounts the same difficulty had appeared at a Black forum during the 2016 campaign, and the criticism then was the same one: that his platform had not been built to center the injustices facing Black Americans in particular. The repetition across two cycles is what lifts the episode above a single bad night.
What is worth watching is not the booing and not the discomfort. It is the shape of the response. Asked something posed specifically — about Black women, about this group's experience of this harm — Sanders reached, under pressure, for the most universal available materials: a decades-old civil-rights credential and the general language of justice for all. He translated a situated question back into universal register almost in real time. That is consistent with the reflex Davis described. It is not proof that her critique is right about his whole career, and it says nothing about his private beliefs. But as a live behavioral sample it is unusually legible.
The disciplined reading must include the possibility that this was simply a poor answer under hostile conditions — that a better-prepared politician with the identical framework would have handled the moment and the question would never have arisen. That possibility is real and should not be waved away. What makes the episode evidence rather than anecdote is not any single stumble but the fact that the same reflex, reaching for the same universal material, appeared twice, years apart, in response to the same kind of specific ask. A reader is entitled to weigh that and decide how much it carries.
VWhere Abstraction Meets Material Power
A debate stage is one kind of test. A policy that would place hazardous material in a specific community is another, and it is heavier. In 1998, then-Representative Sanders co-sponsored and helped move through Congress the compact allowing Vermont and Maine to send their low-level radioactive waste to Texas. The widely expected destination was Sierra Blanca, a town of a few hundred people near the Mexican border — roughly two-thirds Latino, with an average annual income around eight thousand dollars. The waste was low-level, and Sanders had, and has, a long record of opposition to nuclear power. Texas regulators ultimately rejected the Sierra Blanca site in October 1998 on geological grounds, and the waste never went there.
The clearest voice against Sanders's role did not come from a rival campaign. It came from Robert Bullard, the scholar generally credited as the founder of the environmental-justice movement. Bullard's assessment, given to the Texas Tribune, was blunt: it reflected poorly on Sanders, and forcing the arrangement on the community was "not progressive politics" but "business as usual." There is also a contemporaneous quotation, reported by the Texas Observer in September 1998 and treated as authentic by nonpartisan fact-checkers, in which Sanders reportedly declined to visit the site, telling a Texas delegation, "Absolutely not. I'm gonna be running for re-election in the state of Vermont." That quotation rests on a single contemporaneous source and is used here only as reported; the analysis does not depend on it.
Read through this essay's concern, the episode has a precise shape, and it is a shape environmental justice exists to name. From Vermont, the compact is an interstate waste-management arrangement — a reasonable, even responsible, allocation of a burden three states jointly produced. From Sierra Blanca, the same compact is other states exporting a hazardous burden into a poor, majority-Latino border town. Both descriptions can be administratively true at once without being morally equivalent, and the gap between them is exactly the gap a universal frame does not register. Bullard's critique lives in that gap. The pattern here is not that Sanders sought out a Latino town; it is that the arrangement was legible to his framework as an interstate solution, and the town's particular position within it was the part a universal frame is least equipped to weigh.
The limits of this case must be stated as plainly as its force. The waste never arrived; the site was rejected by Texas regulators, not by any reversal on Sanders's part. The material was low-level. And a single episode from 1998, however sharp, is one data point, not a career. It earns its place in this essay for one reason: it is the cleanest available instance in which the translation of a situated harm into a universal administrative frame can be watched happening, named by the field's own founding scholar, in the documented record. It illustrates the mechanism. It does not, by itself, prove it operates everywhere.
VIThe Most Universal Threat Possible
Everything so far concerns how Sanders reads a harm. What makes that reading matter beyond his own record is where it travels next — into an industry whose value depends, in part, on the world believing a specific thing about the technology at its center. To see how the two connect, start with what Sanders himself said he was going into, before he went.
On February 4, 2026, his office announced that he would travel to California to meet AI leaders, and named the four questions he intended to pursue: who controls AI and whether billionaires will monopolize its benefits; what AI and robotics will do to workers; what AI will do to democracy; and whether AI poses an existential threat to the human species. The statement concluded that "AI and robotics are the most transformative technologies in the history of humanity." The existential frame was already his, in his own words, before any meeting. Nobody introduced it to him. He arrived carrying it.
Notice the shape of that list. Every item is pitched at the widest possible scale — billionaires versus everyone, workers as a class, democracy as such, the species. This is the instrument described in Section I, pointed at AI: it reads harm most clearly when the harm is universal. And of the four, the last is the most universal political subject that can be named. Not Black people, not women, not immigrants, not the specific populations that documented algorithmic systems already sort and score and deny — but humanity, entire, at once. If the pattern this essay has traced is real, then a species-level threat is the single most legible AI danger Sanders's framework could possibly encounter. It is almost perfectly shaped for his instrument.
His own writing bears this out. In April 2026, Sanders published a long op-ed, "Artificial intelligence is coming for the working class," laying out the AI-harm landscape in his own priority order. The overwhelming majority of it concerns worker replacement — the Waymo ride, driverless trucks in Texas, Bezos's factories, Musk's robots, the automation of white-collar work. This is his framework at full power, and the analysis is genuinely sharp. He then turns briefly, in a closing run of single "It means that…" clauses, to other harms — children, democracy, privacy — and finally, in the most elevated closing position, to the species-level threat. Those closing harms are themselves largely universal: children as a class, democracy as such, "every aspect of our lives." Even surveillance arrives as something done to everyone equally.
What does not enter the frame at all is the situated category — the uneven distribution of AI harm across particular populations. Algorithmic systems that misidentify Black faces; hiring and lending and risk-scoring tools that fall disproportionately on identifiable groups; sexual deepfakes that overwhelmingly target women; the applications aimed at immigrants and the policed. Across the 2026 record reviewed for this essay — the op-ed, the superintelligence-ban announcement, the data-center moratorium, a September statement on mass surveillance — no such emphasis was found. This is a statement about the reviewed record, not a claim of total absence from everything Sanders has ever said. But the omission is legible, and it is not for lack of an available vocabulary: Alexandria Ocasio-Cortez, Sanders's own ally on the data-center moratorium, spent the same period naming nonconsensual sexual deepfakes and the concrete, present-tense damage AI does to specific people. The situated frame was there, in use, on the same issue, by the person beside him. His own AI material reached past it toward the universal.
Then he arrived in California, and sat down with the people most committed to the most universal frame of all. As this publication documented in The Network Behind the Moderate, Sanders met in Berkeley in February 2026 with Eliezer Yudkowsky and Nate Soares of the Machine Intelligence Research Institute — an organization that received substantial early funding from Peter Thiel — and months later announced legislation to ban artificial superintelligence outright. The discipline that essay established holds here without qualification: there is no evidence of ideological affinity between Sanders and that network, no evidence of Thiel influence over Sanders, no evidence the meeting caused the legislation. Politico called Sanders and Yudkowsky "opposite poles." Exposure followed by congruent escalation is not proof of transmission, and this essay does not claim it. What it examines is a stranger thing that requires no transmission at all.
The stranger thing is what Yudkowsky's own opponents say about him. The conventional story casts AI builders and AI doomers as enemies, and over what should be done, they are: the builder races to construct the thing the doomer wants halted. But listen to their descriptions of the technology rather than their prescriptions, and the enmity dissolves. Yudkowsky's claim is not that frontier AI is overhyped, mediocre, or commercially trivial. His claim is that it may become so extraordinarily capable that everyone dies. The builder says: this technology may transform civilization. The doomer says: yes — so completely that it may end civilization. The prescriptions collide. The premise of extraordinary power is shared. And extreme perceived capability is exactly what is valuable to an industry whose valuation rests on anticipated future capacity.
This is not an inference about how the builders see it. They have said so. In 2023, OpenAI's CEO Sam Altman posted that "eliezer has IMO done more to accelerate AGI than anyone else," crediting him with getting "many of us interested in AGI," helping DeepMind get funded "at a time when AGI was extremely outside the overton window," and being "critical in the decision to start openai" — adding that Yudkowsky might someday "deserve the nobel peace prize for this." Altman's Twitter bio at one point read "eliezer yudkowsky fan fiction account." A 2026 MIT Technology Review feature on how AGI became a consequential belief system quotes the same statement and reports the same chain of influence. Observers split over whether Altman was complimenting Yudkowsky or trolling him, since acceleration is precisely what Yudkowsky regards as catastrophic. That ambiguity is not the point; the private motive can stay unresolved. The documented fact is that the CEO of the industry's most valuable company publicly credited its leading doomsayer with helping make the technology fundable, discussable, and central.
This is where the older Horizon Accord work on belief becomes load-bearing rather than decorative. The Enron Parable named the mechanism: an AI industry that sells "intelligence futures" — valuation tied not to present cash flow but to faith in inevitability, future capacity monetized before it arrives. When the Thing That Bursts Is Belief took the next step: in such a system, it is not the product that inflates the value but the language around it — belief itself circulating as infrastructure. Set the extinction frame inside that machine and its paradoxical function appears. The more powerful AI must be for the catastrophe story to hold, the more consequential the underlying technology is made to seem. A warning pitched at the scale of human extinction cannot help but assert that the thing it warns against is the most important object in the world. Sincere opposition and market-useful significance are, in this one respect, produced by the same sentence.
None of this requires anyone to be lying, coordinating, or captured. The builder can genuinely want to build and the doomer genuinely want to stop him, and their conflict can be total, while both continue to broadcast the shared premise: this technology is unlike ordinary technology; it changes everything; it demands extraordinary attention; it belongs at the center of governance. That premise is political value. It is also, for actors whose capital depends on future capability being believed, economically useful. No secret alliance is needed. Functional convergence is enough.
Which is what makes Sanders matter differently from either of them. Altman's stake in AI's importance is obvious; he runs the company. Yudkowsky has spent decades arguing that AI is important enough to end humanity. Sanders is the anomaly, because his credibility is built on opposing concentrated technological and billionaire power. When he treats species-level AI danger as worthy of a permanent federal ban, the extinction premise acquires a validator whose authority does not originate inside the builder industry or the rationalist ecosystem that produced it. The frame crosses an ideological boundary. It picks up anti-oligarch legitimacy — the one endorsement Silicon Valley cannot manufacture for itself. This does not mean Sanders intends to market anything. It means a warning delivered by a trusted adversary of the industry can still ratify the industry's most valuable claim about itself: that its product is civilization-defining. Fear can constrain a technology while enlarging its significance. That is not a paradox in Sanders's psychology. It is a property of the frame he carried in, sincerely, on his own.
VIIThe Uses and Limits of Universalism
Gather the samples and a shape appears. When a harm can be named as a universal principle, Sanders's framework reads it with rare clarity and he acts, often early and at real cost: DOMA, welfare reform, the PATRIOT Act, Iraq. When a harm requires centering a specifically situated group — Black women asked about as Black women, a Latino border town carrying a burden allocated at the altitude of interstate compacts, populations sorted by algorithms rather than humanity threatened as a whole — the same instrument tends to translate the specific back into the universal, and something falls out in the translation. The strength and the limit are the same instrument, seen from two angles.
If this were only about Sanders, it would be an interesting limitation and little more. It is more than that because of what the specific instrument meets in the present moment. There is, right now, an industry whose financial and political power depends substantially on convincing the world that its technology represents the future of civilization — that its product is not one tool among many but the axis on which everything turns. That industry's most valuable asset is not a model. It is a belief about the model's future capacity. And the frame best suited to enlarge that belief is precisely the one pitched at the largest possible scale: not a harm to some, but a stake for all.
That is the frame Sanders's instrument reads most easily, and the frame he carried into governance. And when he carries it, it does not merely gain a prominent voice. It gains legitimacy from outside the ecosystem that produced it — from a man whose entire public authority rests on refusing to take Silicon Valley at its word. That is a transmission mechanism, not a personal blind spot. A trusted opponent of concentrated technological power can sincerely try to constrain an industry while helping certify that the industry's product is important enough to reorganize the state around. The warning and the advertisement, in this one structural respect, are the same broadcast.
So return to Davis, at the larger scale her critique was always reaching toward. What falls out of a universal frame is not merely specificity. It is the distribution of power itself: who is harmed first and who is protected longest, who is sorted by the algorithm and who writes it, who profits while the danger is being narrated, and whose version of the future becomes important enough to govern. A framework that struggles to hold the specific will struggle to hold exactly those questions — and they are the questions that decide who the political order actually serves. The limitation was never politically neutral. In an economy that runs on belief in extraordinary future capability, it is the most consequential thing about the instrument.
