Old Man Yells at Claude

An email from the Bernie Sanders team landed in my inbox this week. At the bottom was the usual ask — $27, or whatever you can afford — and above it, a full reprint of an op-ed he’d just run in the Times. It’s the one he reads aloud in the video above. The pitch is clean enough to fit on a sticker: the public should own half of every big AI company. A one-time fifty-percent tax, paid not out of profits but in stock, the shares dropped into a sovereign wealth fund that cuts checks to every American. OpenAI, Anthropic, xAI — named directly. I read the whole thing on a laptop running tools built by two of the three.

I expected to skim it, agree or disagree, and get on with my day. Instead it pulled me in, and the thing I came out the other side with is not a thing I can print on either of the bumper stickers this debate ships with.

There’s a standard way to read an op-ed like this, and it’s as a referendum. The left reads it and cheers — finally, the people who built this get a cut. The right reads it and reaches for the words theft and innovation and China. Both sides, underneath the noise, agree on what the question is: should the public own AI, yes or no. Pick a jersey.

I came in already agreeing with Bernie’s diagnosis — more than I agree with almost any politician about almost anything. What took some working out was exactly why I can’t follow him to the cure. And I want to be careful here, because I’m not fence-sitting. It’s a specific, stable position, and the reason it’s stable is structural, and older than AI.

The part Bernie gets right

I went in expecting the thing Bernie is genuinely good at — the 1%-versus-99% move, where a sprawling problem gets compressed into one clean, sayable claim — and he delivers it. The hard part isn’t whether the public has a claim — it’s quantifying it. Your specific tweets and comments are worth approximately nothing to any given model; the marginal value of any one person’s contribution rounds to zero. There’s no honest way to say you’re owed 0.0001% and I’m owed 0.00007%. Attribution, in other words, is hopeless.

But that difficulty matters far less than it looks, for two reasons. The first is scale. A rounding-error share of nothing is nothing; a rounding-error share of a multi-trillion-dollar industry is a check most Americans would actually feel. The size of the prize is exactly what turns those absurd-looking percentages into real money. The second is that distribution doesn’t need attribution. You don’t have to figure out what anyone’s slice was worth to say “the commons made this, so everyone gets an equal share.” That’s just Alaska — no resident has to prove a personal relationship to a barrel of oil; residency is the whole claim. Once you stop trying to individualize it, the hopeless part evaporates and you’re left with a simpler question: is there a real thing here, collectively built, that’s now privately owned?

And there obviously is. The thing that makes the ownership frame fair isn’t some mystical claim on the training data. It’s that the models themselves are concrete, balance-sheet, legally-defended assets of for-profit companies. There’s no philosophical fuzz about whether a model is property — OpenAI and Anthropic will tell you it is, loudly, the second anyone tries to copy one. So the question gets very sharp: a definitely-owned asset was built out of an input nobody paid for. The remix-skeptic in me wants to say “but everyone builds on everyone, culture is theft all the way down” — and that’s a fine argument against individual royalties, but it doesn’t touch the structural fact that there’s a clearly-owned thing here whose existence depended on an unpaid collective input.

The one honest correction — and it’s the thing that keeps the whole argument from tipping into nonsense — is that the data was necessary but nowhere near sufficient. You cannot pile up every book and tweet and get a model out the other end. The capital is staggering: training runs in the nine and ten figures, data centers, and the electricity especially, which is quietly becoming the binding constraint on the entire industry. So the defensible claim was never “the public built this, hand it over.” It’s “the public supplied one essential, irreplaceable input among several expensive ones, and deserves a share proportional to that contribution.” Which is a real claim. It’s also a claim that, taken seriously, does not produce the number fifty. Fifty is rhetoric. The honest version points somewhere smaller and a lot more negotiable — and notice that the more carefully you make Bernie’s own argument, the further it drifts from Bernie’s own number.

Why nobody wrote this op-ed about Linux

Here’s the part that actually made the whole exercise worth it, because it explained why this question feels new.

Nobody ever proposed a sovereign wealth fund for Linux. Nobody argued the public should own a stake in the open-source software that quietly runs every server, phone, and cloud on earth. And the reason isn’t that software is less valuable than AI. It’s that in software, the means of production was a laptop and an internet connection. A teenager with a hand-me-down PC could submit a patch to the kernel, and enough of those patches aggregated into something that runs the world. Participation was practically free, so ownership took care of itself. You owned your piece because you could just go make it.

AI breaks that, and not by anyone’s choice — by physics and economics. You cannot crowdsource a frontier pretraining run the way you crowdsource a codebase, because it has to happen on one tightly-coupled cluster of tens of thousands of accelerators running for months. There’s no “submit your slice from your laptop” for the part that costs the money. The community can do everything that decomposes into small pieces — fine-tuning, evals, tools, apps on top of a released model — but the base model itself has to come down from someone who could afford the data center. The grassroots can decorate the building. Somebody with capital had to pour the foundation.

Which is why “open” in AI is a weaker word than it was in software. Open-source meant you could see the source, change it, and rebuild the whole thing yourself. “Open-weight” usually means you get the finished model to run and tweak, but not the data, not the full recipe, and not the means to reproduce it — because reproducing it costs nine figures. You’re handed the artifact, not the capacity to have made it. That’s useful. It’s also closer to freeware you’re allowed to modify than to anything that earns the word open in the way the last generation meant it.

So here’s the thing that snapped into place: the cost floor is what resurrects the ownership question. When participation was free, ownership dissolved as a problem — you could just join. When participation costs a data-center lease, ordinary people are structurally locked out of the production side, and the only routes left to broad ownership are the ones that involve pooling. A public fund. A sponsor choosing to commoditize the layer. Government provision. Bernie is responding to something real and genuinely new. The lock-out isn’t imagined. It’s load-bearing in his whole case.

What America actually does with a resource

So say you buy the principle, as I did. Then you run into the history, and the history is not encouraging.

The cleanest precedent is Standard Oil. By the 1880s Rockefeller controlled something like ninety percent of American refining — through railroad rebates, through buying or crushing every competitor, through the trust structure itself, which is where we got the word antitrust. It was exactly the resented, dominant, private control of a critical resource that you’d think might trigger public ownership. There was a massive populist backlash, muckraking journalism, real political heat. And the American response was not to take it over. It was the Sherman Act, and ultimately the 1911 ruling that broke Standard Oil into thirty-odd pieces — the ancestors of Exxon, Mobil, Chevron, the rest. The remedy for too-concentrated private power was to shatter it back into competing private firms.

That’s the reflex, and it has held with remarkable consistency. AT&T in ‘84. The attempt on Microsoft in the ’90s. The antitrust scrutiny circling Google and the big labs right now. When America decides a company is too powerful, it reaches for the hammer that restores competition. It does not reach for the public ledger.

The case Bernie actually cites cuts against him for exactly this reason. Norway’s nearly-two-trillion-dollar fund works because the Norwegian state claimed the offshore oil as public before anyone drilled. It never had to take the resource out of private hands, because it owned it from the start. Bernie is trying to run the Norway play in a country whose entire legal and political tradition runs the other direction — where the resource starts private and the most the state has ever reliably done is referee the fight over it.

There’s exactly one American counterexample, and it proves the rule. In 2008 the federal government did take equity — in the banks, in AIG, in GM. So it isn’t literally unheard of. But look at how it was sold: as a temporary, reluctant, hold-your-nose emergency, with everyone swearing they’d unwind it and hand it all back as fast as humanly possible. Which they did. Even at the high-water mark of American state intervention, permanent public ownership was the thing everyone was at pains to insist they weren’t doing. Bernie is proposing as a deliberate, permanent design the one thing the U.S. has only ever tolerated as a crisis-era embarrassment to be escaped quickly.

And underneath the policy sits the culture, which is the part you can’t legislate around. The American story is the self-made one — you build it, you risk it, you own what you made, and property isn’t just an economic arrangement, it’s tangled up with liberty itself. “The public should own half of what you built” doesn’t lose on the merits in that frame. It registers as a moral violation, a kind of theft, before the argument even gets a hearing. For better or worse — and I’m describing this, not defending it — that reflex is in the country’s DNA. You can win a mechanism debate with a better mechanism. You can’t win against a story a country tells about who it is. Norway can do this because Norwegians are working from a different founding myth, one where some things are collectively held by default. The fund is downstream of the culture. You can’t pour the fund in first and hope the culture arrives later.

The asset that moves the fleet

Then there’s the wrinkle Bernie’s framing quietly leaves out, and it’s the one that reorganizes everything above it.

This isn’t consumer technology anymore. It’s a strategic asset, and the oil parallel gets tighter here than anywhere else. Oil was never just a commodity — it moved the fleet and the air force, which is why “energy security” became a category worth almost any cost. Once a resource becomes load-bearing for military power, the state’s relationship to it stops being about consumer welfare or fair distribution and becomes about control, secrecy, and denial to adversaries. AI is crossing that line right now — autonomous systems, intelligence analysis, cyber, logistics, targeting. The big labs are already defense contractors in everything but the letterhead.

And that flips Bernie’s argument inside out. His is a distributive claim: the public built it, the public should share the upside. But the national-security interest doesn’t care about distribution at all. It cares about control and advantage. The government may well end up deeply entangled with these labs — as a customer and a controller, classifying capabilities, restricting access, steering development — and that is the opposite of cutting dividend checks to citizens. Both are “more government in AI.” They point in violently different directions, and history is not kind to the distributive one when it collides with the strategic one.

The genuinely dark part is that the defense relationship tends to protect the incumbents rather than discipline them. You don’t break up the company building your weapons systems. You don’t heavily tax the lab the NSA depends on. The military entanglement is a moat dressed as patriotism, and it shields the labs from exactly the public-interest constraints we’ve been discussing — Bernie’s fund and the antitrust hammer alike. Which is why the most likely future is none of the clean options. It’s nominally-private national champions, fused with the state through classified contracts, too strategically important to tax hard or break apart, with “the public interest” expressed through procurement and industrial policy rather than any dividend to anyone. Neither the capture Bernie fears nor the democratization he wants. A third thing that mostly serves the institutions already at the table.

A control structure doesn’t care why you built it

Which is where the whole thing finally landed, and not comfortably.

You can run a paranoid version of this. Someone could argue that the feel-good lefty framing — the people should own it — is unwitting cover for what the security state already wants anyway, which is a controlling stake in every frontier lab. Bernie has a forty-year record of meaning the redistributive thing literally. But that record is the unsettling part. A control structure doesn’t always get built by the villain everyone sees coming; sometimes it’s built by the most credible person in the room, the one whose sincerity puts the machine beyond suspicion. Bernie’s believability is what makes him the perfect Trojan horse — whether he means to be or not. For the record, I don’t think he does.

But the value of the paranoid version isn’t in believing it. It’s that it stress-tests the structure, and the structure is the only thing that actually matters here. Because a measure that only looks safe when you assume good actors administering it forever is broken by design — you never get good actors forever. The keys always, eventually, change hands.

That’s the whole thing, and it’s why the principle and the mechanism come apart so cleanly. You can hold “the public has a real claim on the upside” and reject “the federal government holds the equity and the board seats,” and there is nothing incoherent about that. The lever outlives the intention that built it. You build the door for one reason; whoever inherits the keys uses it for theirs. “Government voting shares and a seat on the board of every frontier lab” is a control structure regardless of how pure the sentiment in the press release was — and a control structure is dangerous in proportion to its existence, not its motive. The honest rule is the conservative-in-the-old-sense one: don’t build a lever you wouldn’t want your worst future government holding. And that rule cuts against the security-state version and the feel-good version exactly equally, because they build the same machine.

Where this lands

There’s a line of Thomas Sowell’s that I kept circling back to: there are no solutions, only trade-offs.

Bernie’s entire rhetorical job is to make the trade-off disappear. “You should own half” presents as pure upside — free money you were owed all along, no cost named, no price tag attached. Everything above has been me putting the costs back on the table, one at a time: the capital that the data alone can’t supply, the competitive hit of spooking that capital right before these companies need it most, the lever a worse government one day inherits, the defense moat that shields the very firms it claims to check. None of that makes him wrong. It makes him incomplete, in the specific way slogans have to be incomplete. A trade-off acknowledged doesn’t move anybody. A solution proclaimed does. That’s not a knock on him so much as a description of the job he’s actually doing, which is getting “the public has a claim on AI” into circulation as a sayable political sentence so that other people can argue about the architecture.

But the line cuts the other way too, and this is the part I’d want anyone nodding along with me to sit with. Doing nothing is also a trade-off. Letting the upside pool into a handful of hands has a cost — it’s just diffuse and deferred and doesn’t show up as a dramatic policy with a number on it. “No solutions, only trade-offs” is not a license to stand still and feel clever. It’s a demand that every option, including the comfortable status-quo one, show you its price tag.

So here’s the uncomfortable place I actually ended up. The diagnosis I’ll defend without flinching: AI is built on collective work, the upside shouldn’t pool into a few hands, and ordinary people getting structurally locked out of the production side is a real and new problem. It’s the prescription where the whole thing falls apart for me — because his cure runs straight through the one institution I trust least to hold that much power. And there may be no clean way out of that. Sometimes the honest position is the problem is real, and every fix I can see is worse than the problem, and you sit in that discomfort instead of reaching for the one that feels good and quietly builds a lever you’ll regret handing over.

The email’s still in my inbox.