On July 2, 2026, the Financial Times reported that OpenAI CEO Sam Altman had pitched the Trump administration a proposal to hand the US government a 5% equity stake in the company. Based on OpenAI’s $852 billion valuation from its March funding round, that 5% slice would be worth approximately $42.6 billion. Altman has framed the arrangement as a democratic act: giving the American public a financial stake in the AI revolution. The proposal envisions a structure modelled on Alaska’s Permanent Fund, which distributes oil revenue dividends directly to state residents, applied here to AI equity.
It is a compelling pitch. It is also one of the most consequential corporate governance proposals in the history of the technology industry. The implications deserve considerably more scrutiny than the headlines have given them.
What Stakeholder Theory Actually Predicts Here
The foundational tension in this deal runs directly through one of corporate governance’s oldest debates: who are companies actually for?
Milton Friedman’s 1970 answer was unambiguous. A company’s sole social responsibility is to increase its profits for its shareholders. Any other use of corporate resources like philanthropy, social programmes, handing equity to governments is, in Friedman’s framework, a form of theft from shareholders, whose money is being spent without their consent.
R. Edward Freeman’s Stakeholder Theory, developed in the 1980s, offered a direct counter. Companies exist within a web of relationships with shareholders, employees, customers, governments, and society at large and long-term value creation requires managing all of those relationships, not just the one with capital providers.
The OpenAI proposal is, on its surface, the most ambitious application of Freeman’s Stakeholder Theory in tech history. Altman is literally putting a sovereign government into the equity structure, not metaphorically “considering” its interests. He has also proposed that Google, Meta, and Anthropic do the same effectively proposing a sector-wide rewrite of who American AI companies are accountable to.
The problem is that Stakeholder Theory, even in Freeman’s original formulation, was never intended to dissolve the distinction between accountability and ownership. Having a responsibility to consider the government’s interests in your decisions is fundamentally different from giving the government a financial interest in your outcomes. OpenAI is proposing the latter. That distinction matters enormously.
The Regulatory Capture Problem Nobody Is Naming
In political economy, regulatory capture describes what happens when a regulatory agency develops interests that align more closely with the industry it is supposed to oversee than with the public it is supposed to protect. The classic form runs industry-to-regulator: companies lobby, fund, and eventually staff the agencies that write the rules about them, until the regulator becomes an advocate for the regulated.
What OpenAI is proposing is a structural version of capture running in the opposite direction and it is arguably more dangerous. If the US government holds a 5% equity stake in OpenAI, it has a direct financial interest in OpenAI’s success. Every dollar of OpenAI’s valuation growth increases the value of the government’s holding. Every regulatory decision that constrains OpenAI costs the government money.
Consider what that means in practice. The Department of Commerce is currently navigating export controls on advanced AI models. The Treasury is developing frameworks for AI taxation. Congress is debating liability rules for AI-generated content. In each of these policy processes, the US government would now arrive at the table as a financially interested party, not a neutral arbiter. The conflict of interest is not theoretical. It is structural, permanent, and worth approximately $42.6 billion.
The Intel precedent that the Trump administration cites does not actually resolve this problem, it confirms it. When the government took a 10% stake in Intel in exchange for $8.9 billion in chip manufacturing subsidies, it was buying a specific policy outcome with public money. The OpenAI arrangement is structurally different: the government receives equity for free, in exchange for what is implicitly a regulatory accommodation. Intel traded equity for capital. OpenAI is trading equity for goodwill. That is a meaningfully different transaction.
The Agency Theory Problem Sam Altman Cannot Ignore
Agency Theory describes the relationship between principals, those who own or have claims on an organisation and agents, and those hired to manage it on their behalf. The central challenge is ensuring that agents act in principals’ interests rather than their own, particularly when those interests diverge.
OpenAI’s current principal structure is already unusually complex. Microsoft holds a significant equity position. SoftBank led the March funding round. Dozens of institutional investors participated at the $852 billion valuation. Each of these principals has a financial claim on OpenAI’s future and a reasonable expectation that management will act in their interests.
Add the US government as a 5% equity holder and Sam Altman’s agency problem becomes considerably more complicated. A financial investor wants returns: margin expansion, revenue growth, an eventual IPO at a premium. The US government wants something different: AI leadership over China, national security assurance, and at least in Altman’s framing broad public benefit. These goals may align much of the time. They may also diverge sharply at moments that matter most.
Imagine OpenAI developing a capability that would be enormously commercially valuable but raises national security concerns. A financial principal says monetise it. The government principal says restrict it. Altman, as agent, must navigate between them. Currently that tension is managed through regulation, an arms-length process with defined rules. Under the proposed arrangement, the government sits inside the ownership structure and can apply pressure through a channel that bypasses public regulatory process entirely.
That is not a hypothetical concern. It has already happened. Earlier this year, Anthropic was temporarily required to freeze access to its advanced models under an emergency government export directive, a dispute serious enough to trigger federal legal action. If the government had held an equity stake in Anthropic at that moment, the same pressure could have been applied as a shareholder instruction rather than a regulatory order, with none of the procedural transparency that formal regulation requires.
The Alaska Comparison That Does Not Quite Hold
The Alaska Permanent Fund analogy that Altman keeps deploying deserves examination. The fund, established in 1976, takes a portion of the state’s oil revenues and invests them in a diversified portfolio, paying annual dividends to Alaska residents. As of May 31, 2026, it was valued at nearly $91.2 billion.
The analogy is structurally appealing. Oil is a natural resource that belongs to the public; oil companies extract it and pay royalties; the state distributes those royalties to residents. If AI is the new oil, and OpenAI is extracting value from data and human knowledge that arguably belong to the public, then a similar royalty arrangement seems logical.
The problem is that oil companies do not write their own royalty rates, structure their own extraction licences, or negotiate directly with the governor about which regulations apply to them. The Alaska Permanent Fund was built on an arms-length commercial relationship, not on the oil industry offering the state a gift in exchange for regulatory warmth. The structural integrity of the model depends entirely on that separation, which the OpenAI proposal explicitly collapses.
Bernie Sanders has pointed this out, from the left, by arguing that 5% is a watered-down substitute for genuine public ownership. The conservative critique runs in the opposite direction that any government equity stake in private companies is a step toward command economics. Both critiques share an underlying concern: that this arrangement blurs lines that governance structures exist precisely to maintain.
The Ground Floor Take
The public benefit case for this proposal is not imaginary. AI’s gains are currently being distributed almost entirely to capital holders, the investors, founders, and employees of a handful of companies whose combined market capitalisations now exceed the GDP of most nations. A mechanism that routes some of that value to the public, as the Alaska Permanent Fund does with oil, addresses a real and growing legitimacy problem for the AI industry.
But the mechanism Altman is proposing introduces governance risks that the public benefit framing tends to obscure. When the entity responsible for regulating an industry also holds a financial stake in its success, the quality of regulation declines, not necessarily because of corruption, but because of the structural impossibility of objective oversight when you have skin in the game.
The most honest read of this proposal is that it is doing two things simultaneously: advancing a genuine idea about public participation in AI’s upside, and solving a very specific problem Sam Altman has right now, which is that the Trump administration has been applying increasing regulatory pressure on frontier AI labs. The 5% stake is, among other things, a $42.6 billion argument for why Washington should want OpenAI to succeed.
Whether that makes it cynical or merely pragmatic depends on what you think corporate governance is actually for. Freeman would say that managing your relationship with government is exactly what responsible companies do. Friedman would say that handing away shareholder equity without their consent to buy regulatory goodwill is a breach of fiduciary duty. Both of them would agree that the arrangement, as proposed, puts Sam Altman in a position where the answer to “who do you work for?” is genuinely, uncomfortably unclear.
