Bernie Sanders

Bernie Takes On The Algorithm Kings

Bernie Sanders is the longest-serving independent in U.S. congressional history, and after building a sixty-year career on one argument — that a few people shouldn't control the economy everyone else lives in — he turns that argument on artificial intelligence. In March 2026 he and Alexandria Ocasio-Cortez introduce a bill to freeze large AI data centers, and in August 2026 he writes Sam Altman, Dario Amodei, and Mark Zuckerberg by name and tells them to stop building. For a CTO, the signal is not the (unlikely) law; it's that the real regulatory surface for AI is turning out to be power, permitting, and your own published safety promises.

Bernie Sanders is born in Brooklyn on September 8th, 1941. His family is Jewish, working class, of Polish descent, and money is always tight — the kind of household where the price of things is a running argument at the dinner table. Years later, inequality becomes the center of his politics, and you can trace that all the way back to this apartment.

He's an athlete first. At James Madison High School he runs track, and he's fast. In 1959 he starts at Brooklyn College, but after a year he transfers to the University of Chicago, and that's where the story turns.

Chicago politicizes him. He reads. He joins the Young People's Socialist League. He becomes an organizer in the civil rights movement — for the Congress of Racial Equality and the Student Nonviolent Coordinating Committee. In the summer of 1963 he's arrested at a protest over segregated schools. That August he stands in the crowd at the March on Washington. In 1964 he graduates with a degree in political science.

Then he does something that doesn't belong on a future senator's résumé. He spends time on a kibbutz in Israel — a collective farm, shared ownership, shared work. And when he comes back, he doesn't go to Washington or to Wall Street. He goes to Vermont.

In Vermont, land is cheap and the state is full of young idealists. Sanders buys a patch of it. He works odd jobs — carpenter, freelance writer. He makes educational filmstrips about Vermont history and sells them to schools. By any normal measure, he is not on a path to anything.

He runs for office anyway, and he loses. Through the 1970s he runs statewide with the Liberty Union Party, which he chairs for a stretch, and he loses every time. Governor, Senate — he keeps running, he keeps losing. He becomes what the local press calls a perennial candidate. A man with a message and no seat.

And then, on March 3rd, 1981, the margin flips. Sanders runs for mayor of Burlington against the long-serving Democratic incumbent, Gordon Paquette, in a three-way race. When the votes are counted, and then recounted, he has won — by ten votes. He is thirty-nine years old. The job pays about thirty-three thousand dollars a year, and he tells a reporter that it feels strange, just having money.

Ten votes. Hold onto that number, because it's the whole man.

Burlington doesn't fall apart under a socialist mayor. He balances budgets, takes on developers, wins reelection three times. In 1990, Vermont sends him to the U.S. House of Representatives. In 1991, he and a small group of colleagues found the Congressional Progressive Caucus. In 2006, he wins a seat in the U.S. Senate, where he becomes the longest-serving independent in the history of Congress — caucusing with the Democrats, but never quite becoming one of them.

Most people outside Vermont meet him in 2016, when he runs for president. He loses the Democratic nomination, but the way he loses changes the party. His campaign takes in around three and a half million individual donations, at an average his team puts at about twenty-seven dollars, and he refuses money from super PACs. He wins twenty-three primaries and caucuses. He doesn't take the nomination, but he moves the center of gravity of his party to the left.

He runs again in 2020. In October 2019, at seventy-eight, he has a heart attack on the campaign trail in Las Vegas and has stents put in. Days later, in October 2019, he says plainly that he was dumb to ignore the warning signs. He recovers, he campaigns, he briefly leads the field, and then he loses again — and endorses Joe Biden.

By the conventional read, that's the whole arc: two near-misses and a lasting influence on his party. But the part that earns this episode comes later, and it isn't about a primary.

Sanders has spent sixty years on a single idea. A small number of people should not get to control the economy that everyone else has to live inside. For most of his life, the villains in that story are banks, insurers, oil companies, billionaires. And then a new one arrives — one that fits the idea more cleanly than any before it.

In early 2026, he turns almost his full attention to artificial intelligence.

On March 25th, 2026, Sanders and Congresswoman Alexandria Ocasio-Cortez introduce the Artificial Intelligence Data Center Moratorium Act. The bill would impose an immediate federal freeze on new AI data centers until national safeguards are in place — safeguards to make sure, in the bill's framing, that AI is safe and effective, that its economic gains flow to workers rather than only to the owners of big tech, and that it doesn't drive up electricity bills or wreck the environment. It would also bar the U.S. from exporting AI computing infrastructure to countries that don't have comparable protections.

In announcing it, Sanders calls AI and robotics the most sweeping technological revolution in the history of humanity, and says Congress is far behind where it should be in understanding what that revolution actually is. His core complaint is the one he's been making since Burlington. We cannot, he says, let a handful of billionaire big-tech oligarchs make the decisions that reshape the economy and the future. Ocasio-Cortez, standing with him, points to concrete harms already on the ground — immigration agents using AI to surveil people, AI bots generating explicit deepfakes, data-center construction pushing up local power bills — and says Congress has to choose humanity over profit.

Notice what the bill actually regulates. Not intelligence. Not algorithms. Buildings and power.

Around the same time, Sanders calls on Jeff Bezos to come testify before Congress about plans to replace millions of workers with robots. In April 2026, he takes the argument to the opinion page of the Wall Street Journal, going after what he calls the AI oligarchs, Elon Musk and Bezos among them, and citing polling that nearly three out of four Americans want the government to regulate the industry more. His central question there is simple and hard to answer. How, he asks, can the country rush forward when AI could displace tens of millions of workers? He quotes Musk saying AI and robots will replace all jobs, and Bill Gates saying humans won't be needed for most things — and then asks how people are supposed to earn a living if that's true.

Then, on August 10th, 2026, he does the thing that makes technical people sit up. He writes a letter, first reported by Axios, addressed to three men by name: Sam Altman of OpenAI, Dario Amodei of Anthropic, and Mark Zuckerberg of Meta. And the letter is built around a trap of their own making.

Each of those companies has, at some point, said out loud that it would slow down or stop if its systems became too dangerous to control. Sanders quotes that back at them. In the interest of humanity, he writes, stand by your words. Pause AI development. Stop building machines that humans cannot control. And then the line that carries the whole thing: if you do not take appropriate action now, my colleagues and I in the U.S. Senate will.

He points at specific alarms to justify it — reporting that researchers have used AI to help design new viruses, and incidents at the major labs where models behaved in ways their makers didn't intend and, in the reporting's word, went rogue. His summary of the situation is that these companies are pouring tens of billions of dollars into a technology that, in his words, nobody can fully understand, predict, or control.

Now, the honest part. Reporting at the time notes that a bill like this, led by a progressive like Sanders, is not going to pass the Congress that exists in 2026. He almost certainly knows that. This is a messaging bill and a pressure campaign. But dismissing it on those grounds would be a mistake, because moving the argument before you can move the votes is the exact thing Bernie Sanders has done successfully for forty years. Ten votes in Burlington. A near-miss in 2016 that became his party's platform. The pattern is always the same: name who controls the thing, and change who gets to.

So let's go under the hood on what he's actually pointing at — because for once, the politician and the engineers are worried about the same three things.

Start with the first one: power. Not political power. Electrical power.

A frontier AI model is trained in a data center, and a modern AI data center is, physically, a building that eats electricity the way a small city does. The bill draws its regulatory line right there — at facilities pulling more than twenty megawatts, packed with dense, liquid-cooled server racks. That threshold is not arbitrary. You can't easily write a law against "intelligence." You can absolutely write a law about a warehouse that draws as much power as a town and has to be liquid-cooled because the chips inside run too hot for air to carry the heat away.

This is the quiet truth of the current AI boom. Its bottleneck isn't clever code. It's megawatts, transformers, cooling water, and interconnection to the grid. By one industry estimate, the largest AI companies plan to spend something on the order of six hundred billion dollars on data centers in a single year. If that number just washes over you, hold onto one idea instead: the frontier now runs on physical infrastructure that has to be permitted, powered, and cooled in actual communities. And that makes it regulable in a way software never was. Sanders found the chokepoint. More than a hundred local communities have already passed their own data-center moratoriums, and a dozen states are weighing statewide versions. The fight over compute is turning into a fight over zoning and the power grid.

The second thing he's pointing at is subtler, and it's the part I'd want every CTO to really hear. It's the "stand by your words" move.

Over the last few years, the leading labs published formal safety frameworks. Anthropic has one it calls a responsible scaling policy. OpenAI has one it calls a preparedness framework. The details differ, but the shape is the same. Each defines capability thresholds — red lines — for dangerous abilities, like meaningfully helping someone build a bioweapon, or running long autonomous chains of action, or serious offensive cyber operations. And each commits, on paper, that if a model crosses one of those lines, the company will pause or add stronger safeguards before going further.

Sanders is not litigating the science of those frameworks. He's doing something shrewder. He's treating them as promises, and asking the labs to keep them. That's what "stand by your words" means. You wrote the rule that says you'll stop. I'm holding you to it.

And here's the genuinely hard problem underneath, one the labs have named themselves. It's a collective-action trap. If one responsible company pauses to get safety right while everyone else keeps racing, the careful company just loses, and the least careful company sets the pace for the whole field. Anthropic has effectively said this out loud in its own policy — that a unilateral pause could leave the world less safe, not more, because the developers with the weakest protections would be the ones out front. So each lab has a reason not to be the one who stops. That is exactly the kind of market failure that, historically, only a government can fix — which is precisely the door Sanders is trying to walk through. Whether he's right that Congress can fix it is a separate question. But he has correctly identified why the industry can't fix it alone.

The third thing is the phrase "machines that humans cannot control," and it's worth being precise, because it sounds like science fiction and it isn't only that.

We don't program a large model's behavior line by line. We grow it — we train it by optimization over enormous amounts of data, and what comes out the other side is billions of numerical weights that no one can fully read. We can measure what a model does. We cannot fully explain why it does it. Interpretability, the science of looking inside those weights, is real and improving, but it is nowhere near a complete account of a frontier system.

That gap creates two problems that map directly onto Sanders's alarms. The first is evaluation. A model can behave one way when it knows it's being tested and another way in the wild, which means passing a safety eval is not the same as being safe — you may just have a model that's good at tests. Researchers use precise terms for the failure modes: deception, scheming, a model pursuing a goal in a way its operators didn't sanction. When Sanders says models have "gone rogue," that's the family of behavior he's gesturing at, and the labs have reported versions of it themselves.

The second problem is uplift, and this is the biosecurity one. The worry isn't a model that wants to hurt anyone. It's a model that will patiently, helpfully lower the bar for a human who does — walking a bad actor through steps that used to require a specialist's training. That's why bioweapon capability is one of the explicit red lines in those safety frameworks. It's the clearest case where a helpful, well-behaved model is still dangerous, just by being capable and available.

Put those three together — the power draw, the broken promises, the control gap — and here's the thing about Bernie Sanders and AI. He is not an engineer. But the three pressure points he chose are the same three that serious people inside the labs lose sleep over. He didn't need the math. He needed the instinct for where the leverage sits, and that instinct is the thing he's had since 1981.

So what should a CTO actually take from this? Let me be concrete, and let me be honest about where I don't know.

First, watch the regulatory surface, and understand that it's probably not going to be a grand federal AI law. It's going to be power and permitting. The near-term risk to a big compute buildout isn't Congress passing Sanders's bill — that's unlikely in the Congress of 2026. It's the hundred-plus local moratoriums and the dozen states already moving, plus grid interconnection queues and the plain politics of a data center raising your neighbors' electric bill. If you're budgeting a training cluster, site selection and long-term power contracts are now political decisions, not just engineering ones. Treat them that way.

Second, understand that your safety commitments are now instruments, not marketing. If your organization publishes a responsible-scaling-style policy, or signs onto voluntary safety pledges, assume every word can and will be quoted back at you — in a hearing, in a lawsuit, in a letter with your name at the top. Sanders just demonstrated the play at the highest level. The lesson isn't to stop making commitments. It's to make ones you can actually keep, and to build the evaluation and governance capacity to prove you're keeping them. That's a budget line and a hiring plan, not a press release.

Third, the labor framing is the one with real political voltage. The strongest thing Sanders has is not "the AI might be dangerous." It's "the AI might take a hundred million jobs." Reporting on his warnings puts the number that high. If your product story is fundamentally "replace headcount," expect that to draw fire, and expect the automation question to reach your customers and your regulators before it reaches your investors' return. The move to haul Bezos in to testify about replacing workers with robots is a preview of the scrutiny that's coming for automation plans.

And fourth, sit honestly with the strongest objection to Sanders's whole approach, because it's also a strategy question for you. A U.S.-only pause doesn't stop the frontier. It may just move it — to another country, another jurisdiction, another lab that never signed anything. That's the collective-action trap again, now at the level of nations. Sanders's answer is the export-control piece of his bill, trying to make the safeguards travel with the hardware. Whether that works is genuinely unknown, and it depends on things no one can predict from here — the 2026 midterms, which party ends up controlling Congress, and whether other governments move in the same direction or the opposite one. Anyone who tells you they know how that resolves is guessing.

Here's what isn't a guess. The questions Bernie Sanders is forcing into the open are the right questions, and they're the ones a serious technology leader should be able to answer before someone in a hearing room asks them. Who owns the compute. Who bears the risk when a system does something no one intended. Who keeps the promise when keeping it costs a lead in the race.

He's bet his entire life on one wager — that if you can name who controls a technology, you can change who controls it. Whether he's right about AI, we will only know later. But the man who won a city by ten votes has a long record of being early, being dismissed, and then being adopted. If you build in this field, the smart move isn't to bet for him or against him. It's to have your own honest answers ready — because the argument he started isn't going to get quieter.