Not a New Deal

Not a New Deal

On 6 April 2026, OpenAI dropped a thirteen-page document into the middle of an already feverish policy conversation and called it a starting point. Its title, "Industrial Policy for the Intelligence Age: Ideas to keep people first," carried the hush of something self-consciously historic.

Sam Altman, the company's chief executive, took to the airwaves and to his preferred medium of long, declarative blog posts to argue that the moment now demanded a new social contract on the scale of the Progressive Era and the New Deal. The proposals inside were the kind of ideas that, only a few years ago, would have made any Silicon Valley boardroom shudder:

• Robot taxes
• A nationally managed public wealth fund seeded in part by AI companies themselves
• Auto-triggering safety nets that activate when displacement metrics cross preset thresholds
• A four-day work week financed by efficiency dividends
• A reorientation of the federal tax base away from payroll and toward capital gains and corporate income, on the grounds that AI will hollow out the wages that fund Social Security

It is, on its face, an extraordinary set of admissions. The company that has done more than any other to accelerate the present wave of labour disruption is now publicly conceding that the disruption is real, that it is large, that it cannot be left to the market to absorb, and that the welfare state as currently constituted will not survive the next decade without significant intervention.

Coming from a firm valued at multiples that depend on continuing to deploy precisely the systems causing the disruption, the document reads less like a policy white paper and more like a confession with a list of conditions attached. The Axios newsletter that broke the story gave it a fitting name. Behind the curtain, this was Sam's superintelligence New Deal.

The framing matters. Franklin Roosevelt's New Deal was negotiated by an elected president and a Congress responding to a Great Depression that no private actor had volunteered to fix. The terms were set by the public, through its representatives, and imposed upon capital. Altman's New Deal arrives in a different order. Capital is at the table first. The terms are being drafted by the entity with the most to gain from a particular shape of settlement. The public, in this telling, is invited to refine, challenge, or choose among the proposals through what OpenAI describes as the democratic process.

Which raises the question that the document itself cannot answer. When the company engineering the disruption is also authoring the response, is the social contract that emerges meaningfully different from one negotiated by the public it affects? And if it is different, in what direction does the difference run?

The Document Itself

The blueprint sets out three stated goals:

1. Distributing the prosperity of AI-driven growth broadly
2. Mitigating the risks associated with superintelligence
3. Democratising access to AI systems and to the broader AI economy

Each is the kind of phrase that has appeared in industry governance literature since ChatGPT's launch in November 2022, and each has the soft, familiar texture of a press release that has been workshopped through several rounds of communications review.

The mechanisms proposed underneath are sharper:

• The public wealth fund would give every American citizen a direct stake in AI-driven economic growth through a nationally managed vehicle that could invest in diversified, long-term assets capturing growth in both AI companies and the broader set of firms adopting and deploying AI. Seed capital would come, in part, from AI companies themselves.
• The automation taxes are described as taxes related to automated labour, with the explicit acknowledgement that the existing payroll-based revenue base cannot survive a transition to capital-intensive production.
• The auto-triggering safety net would scale unemployment benefits, wage insurance, and cash assistance upward as displacement indicators worsen, then phase the supports out as conditions stabilise.
• The four-day work week is presented not as a mandate but as a framework for employers and unions to use efficiency dividends to compress hours without compressing pay.

There are also sections on cyber and biological risks, which Altman has cited as the two most immediate threats from advanced systems, and on the need for a national industrial strategy to keep frontier model development inside the United States. These sit slightly oddly next to the labour and welfare proposals, although they share a common architecture. They are framed as urgent, as inevitable, and as requiring significant public investment in a direction that happens to align with OpenAI's commercial interests.

That alignment is not necessarily a mark against the substance of any individual proposal. A public wealth fund is a serious idea with a long intellectual history, from Norway's sovereign wealth model to the Alaska Permanent Fund to the academic work of economists like Anthony Atkinson. A four-day work week has been trialled in the United Kingdom, Iceland, and Spain with broadly positive results on productivity and worker wellbeing. Robot taxes have been debated since Bill Gates floated the idea in a 2017 interview with Quartz. Auto-triggering fiscal supports were a central feature of pandemic-era proposals from economists across the political spectrum.

None of this is invented from nothing, and the document is careful to nod toward the lineage. What is new is the source. These ideas, when they have appeared in the policy literature before, have come from think tanks, academics, trade unions, and the political left. They have not, as a rule, come from the firms whose business models would be most directly taxed by them.

The sight of OpenAI publishing a blueprint that asks for higher capital gains taxes on people like Altman himself is genuinely unusual. Fortune drew the obvious comparison to JPMorgan Chase chief executive Jamie Dimon, who has periodically called for higher taxes on the wealthy as part of a broader argument about social stability. The intellectual honesty in both cases is real. So is the strategic logic.

The Strategic Logic of Pre-emptive Reform

There is a long tradition in political economy of capital-intensive industries authoring the rules that govern them. Standard Oil did it with the Interstate Commerce Commission. The major broadcasters did it with the Federal Communications Commission. Wall Street did it with vast tracts of the Dodd-Frank legislation. The pattern is well documented in the regulatory capture literature, most influentially by the late economist George Stigler in the 1970s, and the rationale is straightforward. When disruption is coming for an industry, or when the industry is causing disruption that threatens to provoke a public backlash, it is far better to be inside the room where the response is being drafted than to be the subject of someone else's draft.

OpenAI's blueprint fits this pattern with unusual precision. The labour disruption that Altman is now publicly acknowledging is not a hypothetical. It is already showing up in entry-level white-collar hiring data, in the contraction of contract translation work, in the restructuring of customer service operations, in the visible distress of junior coders and graphic designers and copywriters whose work has been automated faster than the labour market can absorb the displacement.

By 2026 the political pressure for some form of response was already building. Unions had begun organising around AI displacement clauses in collective agreements. State legislatures had introduced bills targeting automated decision systems in hiring, lending, and benefits adjudication. The European Union had passed and then partially walked back, through the Digital Omnibus, several sections of the AI Act under industry pressure. The political ground was moving, and the question for any frontier AI lab was no longer whether there would be a regulatory response but what shape it would take.

In that context, getting in front of the conversation with a comprehensive blueprint is exactly what a sophisticated political operator would do. The document does several things at once:

• It signals seriousness, which inoculates against accusations of indifference
• It frames the problem in terms that the company can live with, particularly the assumption that the underlying technology will continue to be developed and deployed at the current pace by the current players
• It offers concessions on tax and welfare that are real but bounded, and that can be negotiated downward as the legislative process unfolds
• It positions Altman personally as a statesman rather than a technologist, which has been a consistent feature of his public posture since the Senate testimony of May 2023
• It shifts the burden of proof onto critics who must now explain why the company's preferred solutions are insufficient, rather than arguing from scratch about whether any solutions are needed at all

The critics noticed. Within hours of the blueprint's release, several prominent voices in AI policy were arguing that the document was a sophisticated exercise in what one called regulatory nihilism. The phrase, picked up by Fortune in its coverage, captures a particular concern. By proposing a vast and ambitious package of reforms that would require years of political work to enact, OpenAI was effectively pushing the response off into the indefinite future while continuing to deploy systems whose effects would compound in the meantime. The blueprint's own language about being a starting point for discussion was, in this reading, a way of ensuring that the discussion never quite reached a conclusion.

There is a more charitable interpretation, and it deserves to be taken seriously. Altman and his colleagues may genuinely believe that the labour transition ahead is severe enough to require something like the New Deal, and that the political system as currently constituted is unlikely to produce such a response without significant prompting from the companies closest to the technology. On this reading, the blueprint is an attempt to use the company's platform and credibility to move a conversation that would otherwise drift. That this also happens to align with OpenAI's commercial interests is a feature, not a bug, because the alignment is what makes the proposal credible to other actors in the room. A blueprint authored by a hostile party could be dismissed. A blueprint authored by the company being asked to pay the new taxes is harder to ignore.

Both interpretations can be true at the same time. The history of progressive reform is full of cases where commercial self-interest and public interest converged on the same policy, and where the resulting legislation was better than either could have produced alone. The New Deal itself was negotiated with significant input from sympathetic capitalists who saw stabilisation as essential to their long-term interests. The question is not whether private interest is involved in public policy, because it always is, but whether the structure of the conversation allows other interests to enter on equal terms.

Who Is Not in the Room

This is where the analogy to the historical New Deal begins to strain. Roosevelt's coalition was assembled from organised labour, urban political machines, agrarian populists, civil rights activists, social workers, and reform-minded intellectuals as well as sympathetic business figures. The Wagner Act, which guaranteed the right to organise, was fought through Congress over the explicit objections of most of American industry. The Social Security Act was drafted by a committee that included the labour secretary Frances Perkins, the first woman to hold a cabinet position, and her staff of social insurance experts, many of whom had spent their careers studying European welfare systems. The terms were set by the public side of the negotiation and the private side accepted them because the alternative, in the depths of the Depression, was worse.

The OpenAI blueprint enters a very different room. There is no equivalent labour movement at the table, because the workers most affected by AI displacement are scattered across freelance markets and white-collar professions that have historically been weakly organised. There is no equivalent agrarian populism, although there are stirrings of an anti-AI politics in rural and small-town America driven by data centre siting disputes and energy costs. There is no Frances Perkins, no figure inside the federal government with both the expertise and the political authority to draft an alternative blueprint from the public side.

The Biden-era executive order on AI was rescinded in January 2025. The current administration's approach has been characterised by a mix of industrial policy support for domestic frontier labs and a general scepticism of regulation. State-level initiatives like California's SB 53 have faced what critics have described as intimidation campaigns from industry, including, by some accounts, from OpenAI itself.

Into that vacuum, the blueprint arrives with the structural advantage of being the only fully developed document in the room. Other actors will respond, and the response will shape the eventual outcome, but they will be responding to a frame that OpenAI has already set. The choice of which proposals to discuss, which mechanisms to specify, which thresholds to use for the auto-triggering safety net, which assets to include in the public wealth fund - all of these have been pre-decided in ways that will be very difficult to undo as the conversation moves forward.

This is the agenda-setting power that political scientists have studied for decades, and it is one of the most consequential forms of influence in any policy debate. The party that writes the first draft almost always wins more than the party that responds to it. The democratic process that OpenAI has invited to refine its proposals will operate within boundaries that the company has already drawn.