Five per cent

In April OpenAI published a policy paper, Industrial Policy for the Intelligence Age, with a proposal. "Create a Public Wealth Fund that provides every citizen, including those not invested in financial markets, with a stake in AI-driven economic growth." The fund "could invest in diversified, long-term assets that capture growth in both AI companies and the broader set of firms adopting and deploying AI", and "returns from the Fund could be distributed directly to citizens". On seeding: "Policymakers and AI companies should work together to determine how to best seed the Fund."

On 2 July the Guardian, carrying a Financial Times scoop, reported what seeding might mean. Sam Altman was in early talks to give the United States government a 5 per cent stake in OpenAI, with other large labs invited to follow, into an investment vehicle "such as the Alaska Permanent Fund". The talks were "conceptual", and any deal could need an act of Congress. The same report has Altman discussing it with the president, two cabinet secretaries and Senator Bernie Sanders, and records Sanders' own proposal: a fund financed by a one-off 50 per cent tax on the largest AI companies' stock.

The same report says OpenAI and Anthropic are preparing listings that some investors think could value each above a trillion dollars, and that both have suggested in their own policy papers that a public or sovereign wealth fund may be needed.

The claim this entry scores comes from chapters 7 and 11: predistribution proposals will keep surfacing as the gains concentrate, most will fail the book's four design criteria, and, in every case the book catalogues, they are demanded of the owners of the machines by people outside them. Chapter 11's politics is a politics of asking. The two proposals the ledger has scored so far, a House bill and Bill Gates's essay, written as a philanthropist proposing a levy on other firms rather than equity in his own, each met one criterion.

Funded from assets, not taxes: meets, on paper. The instrument is equity, where every proposal scored before reached for a levy. It is the mechanism chapter 7 catalogues as Bruenig's, equity paid into a fund in lieu of tax, which the chapter names as the book's preferred family, here offered by a company. Varoufakis's version, a fixed slice of every share issue paid into citizens' own accounts, is the nearest relative; a grant of existing shares to a government fund is Bruenig's shape, not his, though Bruenig's is compulsory and continuing, not a one-off gift.

Universal, not gatekept: meets in its wording, hedged in its verb. "Every citizen, including those not invested in financial markets", with returns that "could be distributed directly to citizens".

Insulated from raid: unbuilt. The paper leaves seeding to "policymakers and AI companies", the July report leaves the vehicle to Congress, and no rule, board or lock has been described. Five per cent of a company that has not listed is a promise on a promise.

Built before the gains are gone: fails. The offer comes ahead of a listing that would set the company's value in the region of a trillion dollars. Five per cent conceded at that point is real value, but the other 95 per cent of the climb from nothing to a trillion stays where it is, and the criterion asks whether the public was in before that climb, not whether it is let in once the company is built.

Two of four, on paper. The sceptic's reading first. A proposal that fails two criteria, exists in no statute, names no vehicle and was floated to an administration in a summer of pressure is a trial balloon, and the claim says most proposals will fail. On the criteria alone, this one confirms it.

What complicates the claim is which two it passes, and who passes them. Both proposals scored before failed by reaching for the dividend and not the deed: a levy on tokens, a tax on robots. This one reaches for the deed. It is the first to name equity as the instrument, and the first to do so is the company.

The second is who makes it. This one comes from the owner of the asset, and the Guardian's first sentence gives the context: AI companies "attempt to smooth relations" with the administration, weeks after, as the same report records, the government ordered one lab to curtail foreign access to its newest model. A stake offered in that setting has a price, and the price is not written in the four criteria. The criteria ask whether a stake will hold. They do not ask what the owner bought by granting it.

The third is the Norway problem. A stake held by the government is a state asset, and chapter 8's finding applies: a possession the citizen never feels can be quietly redirected. Five per cent of OpenAI in a Treasury account is the Alaska principal without the Alaska cheque, which the paper's "could be distributed" leaves open.

Read the offer as what it is: the most criteria-compliant proposal the ledger has seen, made by the one party the criteria never expected to make it, at the moment it costs least and buys most. The book said the deed was the thing to ask for. It did not say what happens when the owner offers it first. Who owns the machine?