In April 2026, OpenAI released a thirteen-page document titled Industrial Policy for the Intelligence Age: Ideas to Keep People First. Six weeks later, in an interview with The Atlantic's Nicholas Thompson, Sam Altman returned to the same themes — and this time the language was sharper. He said plainly that he no longer believed in universal basic income as much as he once did. Cash payments, he said, "do not get at what we're really going to need for this next phase." What people would actually want, in his framing, is "an ownership share in whatever the AI creates" — a claim on something that compounds, that participates in growth, that resembles, in his exact phrase, "universal basic wealth."
It is worth pausing on the vocabulary. Ownership rather than access. Democratic distribution of upside. The right to AI as utility. Worker voice in deployment. Capital-gains taxation paired with shares transferred to a public fund. This is not the language of the Sam Altman who, nearly a decade ago, funded a fourteen-million-dollar universal basic income experiment. It is closer, in surface form, to the language this project has been arguing for. And that is the first thing to register clearly: the most prominent CEO in the field has now adopted the vocabulary of democratic ownership.
The proposals are real, on the page. The paper calls for a Public Wealth Fund "seeded by governments and AI companies" that would give every citizen, including those not invested in financial markets, a stake in AI-driven economic growth. It proposes a "Right to AI" treating affordable access as comparable to literacy, electricity, and the internet. It calls for automated labor taxes — a so-called robot tax — to fund worker transitions. It proposes portable benefits decoupled from employment, formal worker voice in AI deployment decisions, and higher capital gains taxes. Altman has elsewhere proposed seeding the fund by taxing companies above a certain valuation at 2.5% of their market value each year, payable in shares transferred to the fund. This last detail has a Henry-George flavor: a tax on the value of the franchise itself, not just on profits.
If you read the paper alone, with no context, you might conclude that something has shifted at the heart of corporate AI.
But the paper does not exist alone. It exists alongside OpenAI's actual lobbying record, and that record tells a different story.
In 2024, California's State Bill 1047 proposed third-party audits of frontier models, incident reporting, safety protocols before deployment, whistleblower protections, and a public cloud computing cluster — CalCompute — for researchers and startups without access to frontier infrastructure. CalCompute was, in nearly every meaningful respect, the "Right to AI" that OpenAI's 2026 paper now claims to support. OpenAI helped kill it. The bill was vetoed in September 2024 after a sustained industry campaign against its binding provisions.
In Brussels, OpenAI lobbied against provisions of the EU AI Act that would have created greater oversight of high-risk systems. Writing in Tech Policy Press, Paul Nemitz — recently retired Principal Advisor and Director at the European Commission — observes that what OpenAI presents as a visionary, democratically-minded policy agenda is "better understood as a sophisticated exercise in corporate reputation management, designed to support the claim that the company takes democracy and corporate citizenship with a public-interest orientation seriously while it actually preempts and shapes the regulatory environment to its own ends."
The paper invokes the Progressive Era and the New Deal, suggesting that the transition to superintelligence requires an even more ambitious form of industrial policy. Nemitz's response is worth quoting at length:
OpenAI's proposals, by contrast, are largely market-friendly and non-binding — a "starting point for discussion," accompanied by hundred-thousand-dollar grants and a million dollars in API credits for researchers willing to develop the agenda further within OpenAI's frame.
There is also the matter of OpenAI's own internal record. The superalignment team, co-led by Ilya Sutskever and Jan Leike, was promised twenty percent of OpenAI's compute to address safety risks. It received between one and two percent, on aging hardware, before the team was dissolved. Leike resigned publicly, citing what he described as the abandonment of the alignment mission. The "Mission Alignment" team that replaced it lasted sixteen months. The April 2026 paper now asks the public sector to fund safety research — research the company once promised to fund itself, then defunded, then dissolved.
This is the gap the tag at the top of this entry is meant to name. Ownership Rhetoric, Enclosure Practice. The vocabulary of democratic ownership is being borrowed; the binding mechanisms that would deliver any of it are being lobbied against in the same season. The Progressive Era reforms were not opposed by their beneficiaries because they were too modest. They were opposed by industrial firms because they took something. OpenAI's proposals, as Nemitz notes, take nothing — until binding mechanisms appear, at which point OpenAI opposes them.
But the more important objection — the one this project has been making from the start — is structural, and it does not depend on questioning OpenAI's good faith.
Even granting every word of the policy paper its most generous reading, what is being proposed is not democratic ownership of AI. It is a Public Wealth Fund that distributes dividends. The model is closer to Norway's sovereign wealth fund than to anything cooperative. Norwegians get a share of the returns from the country's oil; the oil companies and the state still run the operations. The citizens are stakeholders in returns, not in production. They cannot vote on what gets drilled, where, by whom, under what conditions. They receive a check.
Mondragón is not a Public Wealth Fund. Cooperative Home Care Associates is not a Public Wealth Fund. The Preston Model is not a Public Wealth Fund. These institutions, each in their own register, do something categorically different: they place ownership of productive capacity in the hands of the people who do the producing. The workers who answer the calls, build the bus, teach the children, write the code, do the data labeling, drive the truck — they hold the equity, sit on the board, share the profits, and, crucially, decide together what gets made and how.
The Public Wealth Fund holds none of that. It is a redistribution mechanism layered on top of an economy that remains structured exactly as before. OpenAI continues to make every decision that matters about what gets built, who gets paid, what data gets used, which workers' jobs are automated, and on what timeline. The citizens — recast as fund participants — collect a dividend on someone else's decisions.
This is the difference between the people's share and the people's allowance.
The yeoman tradition that Jefferson romanticized was, for all its exclusions, an argument about the structure of production: a republic of independent producers who owned the tools and the land they worked. The cooperative tradition that grew out of nineteenth-century artisan mutualism, that took form at Mondragón in the wake of the Spanish Civil War, that runs through the kibbutzim and the credit unions and the worker-owned bookstores — that tradition keeps the yeoman question alive while jettisoning the agrarian particulars. Who owns the means? Not who collects the returns.
OpenAI's Industrial Policy answers the second question. This project, and the people whose work it tries to hold, are still asking the first.