AI Does Not Belong to America
AI Does Not Belong to America
Which Public Owns the Machine?
On 1 June 2026, Bernie Sanders made a claim that was moral and material. Artificial intelligence, he wrote, was built from “our collective intelligence”: books, songs, journalism, code, research, conversations and ideas produced across generations. Because the technology rests on this inheritance, the wealth it creates should benefit humanity. Seventeen days later he introduced the American A.I. Sovereign Wealth Fund Act.
The bill would require covered AI companies to issue enough new shares for the United States Treasury to hold half of every class of their equity, and it would authorise Congress to appropriate each year an amount equal to five per cent of the fund’s average market value. Humanity supplies the premise. The American state receives the asset.
That change does not by itself make the proposal illegitimate. States tax income earned within their jurisdiction and spend the proceeds on residents. A jurisdictional defence of Sanders would point to American law, capital markets, public research, procurement, infrastructure and access to a large market, and it would argue that the United States has a serious claim on the rents earned within this system.
The difficulty lies elsewhere. Sanders does not justify the fund only through American jurisdiction or public investment. He invokes the knowledge of humanity as the source of the entitlement. His argument begins with a universal creditor and ends with a national beneficiary, but it does not explain the passage between them.
The omission matters because “the public” is not a natural object waiting to collect its share. A public is made by law. It has borders, membership rules, institutions and the power to exclude. When Sanders says that AI belongs to the public, he seems to name humanity, but when the bill acts, the public becomes the United States Treasury, a federal commission and people inside the United States.
The word carries the warmth of universality and the machinery of citizenship. That ambiguity hides the proposal’s first political decision. The bill directs future appropriations towards payments and services inside one country, not towards an international pool representing the humanity invoked in its moral case.
Knowledge is a difficult resource on which to build a title. Oil lies beneath territory. A government can identify the field, license extraction and collect a rent. Knowledge does not have that shape. A theorem can be used in many countries at once. A language has no single owner. A novel may be protected by copyright, an old text may belong to the public domain, a software library may be open under conditions, and a scientific result may have been publicly funded. These things can all contribute to an AI system, but they do not enter it under one legal status or one moral claim.
“Collective intelligence” therefore describes dependence better than ownership. It tells us that no model company created the world from which it learned. It does not tell us that humanity owns every input in common, still less that one government may convert that inheritance into half the equity of selected firms. Sanders is right to attack the fantasy of the solitary inventor. The model, the chip, the data centre and the firm all stand on institutions and knowledge built by others. But dependence is not yet a title deed. A moral account of origins must still name the claimant, the debt and the rule of payment.
The same difficulty appears when the proposal is treated as restitution. Some training material may have been licensed, some may have been publicly available, some may have been protected by copyright, some may have entered through contested uses, and some may have been produced under open terms. A national equity fund cannot decide which work was used, whether a right was infringed or who should receive compensation. It may capture wealth from firms, but it does not repay the person whose right was violated. A poet in Kenya, a photographer in France and a programmer in Brazil do not become compensated because American residents receive a public dividend.
The strongest defence of Sanders begins where this objection ends. Exact attribution may be impossible. Frontier systems combine vast bodies of data, software dependencies, human labour, compute, energy, capital and organisation, and even a perfect record of provenance would not tell us how much of a company’s value came from one book, one dataset or one laboratory. Contributions overlap, reinforce one another and often matter only in combination. A demand for exact accounts can therefore become a refined way of protecting the status quo. While philosophers search for the proper denominator, shareholders keep the asset.
A national fund can be defended as an instrument of power rather than an account of perfect justice. The United States can reach companies that operate within its law. It can capture part of the upside before private ownership hardens further, and it can create a public shareholder capable of challenging concentrated corporate control.
This is a stronger argument than the claim that American citizens are the rightful heirs of humanity’s knowledge. It is also narrower. It says: this state has jurisdiction, these firms depend on public systems, and democratic government should capture part of the rent. It does not say that the distribution settles the global debt invoked in the preamble.
Yet even this narrower claim opens another problem. Other governments could make the same argument. Europe could claim the value supplied by its users, universities and cultural archives. India could claim the value of its engineers, languages and public digital infrastructure. African states could claim the value of data extracted from markets where firms pay little tax. Countries that host data centres could claim a return for electricity, water and land. Countries that fabricate chips or refine critical minerals could claim that AI wealth begins in their factories and mines. Once value is understood as socially produced, the argument does not stop at one border. It multiplies.
The result might be a more just international settlement, but it might also become a contest among states to enclose different parts of the same knowledge system. Wealthy countries would possess the tax authorities, courts and capital markets needed to collect. Poorer countries might supply labour, minerals, language and data while lacking the institutions needed to turn contribution into ownership. A national wealth fund could therefore correct one concentration of power while reproducing another at the level of states.
The bill itself is not vague rhetoric. It sets a revenue threshold for specified AI activities, requires newly issued shares rather than cash, imposes further share taxes after later issuances to prevent dilution, creates a seven-member commission, and orders mixed AI and non-AI companies to separate their structures. The commission would exercise voting rights and place representatives on company boards while pursuing worker welfare, public safety, fair competition, environmental sustainability and financial solvency.
That provision reveals the proposal’s philosophical interest. The fund is not merely a savings account. It places political purposes inside corporate ownership. Yet those purposes do not naturally agree. A commission may restrain a profitable product because it threatens workers or public safety, and the same decision may reduce the market value on which the public payout is calculated. The state would ask the fund to discipline the companies and depend on their rising valuations. A public owner that takes its social mandate seriously may earn less than a passive investor, while one that protects the dividend may become timid before the firms it was created to govern.
This is not an incidental design flaw. It touches the old problem of public ownership itself. Ownership does not remove conflict. It changes where conflict occurs. Under private ownership, shareholders pursue returns and the state regulates from outside. Under Sanders’s model, the state becomes both owner and regulator, beneficiary and judge. The argument moves from whether government should interfere with the company to which purpose inside government should prevail. Public ownership can widen democratic power, but it can also blur accountability if every institution carries several incompatible duties.
The cash promise contains a similar uncertainty. The bill authorises appropriations equal to five per cent of average market value, but market value is not income. A private company can rise in valuation while paying no dividend, and an unrealised gain cannot finance a cheque unless somebody turns it into cash. The fund could rely on dividends, realised gains, selected asset sales or separate fiscal support, but each route changes the institution. The advertised dividend is possible, but it is not self-executing. The law supplies a valuation formula before it supplies a complete payment mechanism.
The comparison with established sovereign wealth funds sharpens the difference. Alaska distributes part of the return from a fund built on resource revenues. Norway placed petroleum income in a diversified global portfolio spread across many assets and countries. Sanders would begin with a concentrated holding in one volatile sector and give the same public institution financial exposure and governance power over competing firms. The phrase “sovereign wealth fund” joins these arrangements, but their balance sheets, risks and political purposes differ.
Sanders has nevertheless named a real problem. Private firms can turn public research, cultural production, infrastructure and accumulated knowledge into assets of extraordinary scale. Regulation that leaves ownership untouched may leave the central power untouched. Distribution begins before income appears, at the moment when law decides who owns the machine.
But his proposal is overfull. It asks one fund to collect monopoly rent, repay creators, protect workers, govern corporate conduct, finance public services and embody democratic control. These are related debts, but they are not the same debt. Copyright points towards rights holders and licensing. Labour displacement points towards wages, bargaining power and social insurance. Public subsidies and infrastructure point towards taxes, royalties, warrants or revenue shares. Concentrated corporate power points towards competition law, governance and sometimes public ownership. One asset cannot make these claimants identical merely because each has been denied something by the same industry.
A stronger settlement may need several institutions rather than one dramatic fund. Government could tax excess profits and monopoly rents. It could take equity or warrants when companies receive public subsidies, procurement contracts or privileged access to infrastructure. It could require collective licensing where creators can be identified. It could finance wage insurance, shorter working hours and stronger bargaining rights where automation displaces labour. It could support open-source systems, public compute and non-commercial research so that public ownership does not depend entirely on buying a stake in private firms after they have become dominant.
Such a settlement would appear less simple. It would also make the claims clearer. The creator would receive payment as a creator. The worker would receive protection as a worker. The taxpayer would receive a return on public investment. The citizen would gain public services through democratic budgeting. The state would regulate safety and competition without pretending that a single shareholding can perform every function at once.
The border problem remains, but it should be stated carefully. The United States does not need permission from humanity to tax companies within its reach and distribute the proceeds at home. It does need a better argument if it claims that humanity’s inheritance makes Americans the exclusive owners of the resulting fund.
Sanders can narrow the premise and defend the measure as an American claim on rents produced under American institutions. Or he can keep the universal premise and ask how part of the return might support global public goods, cross-border creator payments, open knowledge and neglected language communities. What he cannot do without explanation is use humanity to establish the debt and citizenship to close the account.
The proposal’s deepest achievement may be that it forces the ownership question into public view. Its deepest weakness is that it answers too quickly.
Before deciding that the public should own half of AI, we have to decide which public, on what ground, for which debt and through which institution.
The percentage comes later.