Jennifer Huddleston and Tad DeHaven
Artificial Intelligence (AI) policy shouldn’t begin with the presumption that an emerging technology requires new forms of government control. In fact, the history of American technology policy shows that a light-touch approach allows consumers and innovators to find the best uses for technology. The light-touch approach enables companies to respond to the problems and demands of their consumers rather than those of the government, helping American companies become industry leaders.
Yet concerningly, a new bad policy idea intended to support American leadership in AI is emerging on both the Left and the Right. President Trump has floated a possible federal “partnership” with major AI companies, in which the public could receive “pieces” of those companies and benefit from their success. The details are unclear, but all signs point to the administration seeking to acquire equity stakes, which it has done with over 20 companies starting last year.
On the Left, Senator Bernie Sanders has been more explicit. His proposed American AI Sovereign Wealth Fund Act would impose a one-time 50 percent tax on the largest AI companies, paid in stock. The government would then use voting shares and board representation to block decisions it deemed harmful and push decisions it deemed beneficial.
These proposals come from different political instincts. Sanders wants redistribution and government control. Trump appears more interested in dealmaking and public buy-in. Both risk transforming AI policy from a framework for innovation into a vehicle for state control.
This convergence didn’t come from nowhere. As Tad recently argued, Trump’s equity stake agenda helped create the precedent that Sanders is now trying to use more openly. Once the federal government starts taking ownership stakes in private companies, the remaining fight is over who gets to exercise the power and toward what ends.
Legal Concerns
Sanders’s proposal raises the clearest legal concerns. As our colleague Ilya Somin argues, forcing companies to hand over half their stock would likely run afoul of the Fifth Amendment Takings Clause, which prohibits the government from taking private property without paying just compensation. Labeling the transfer a “tax” doesn’t eliminate the problem, as “Stock is private property, and seizing 50% of the stock value of major firms is a pretty obvious case of confiscation.”
Trump’s intentions are harder to evaluate because they apparently haven’t been fleshed out. But a “voluntary” equity deal with companies subject to federal contracts, antitrust scrutiny, energy and data center policies, and other regulatory approvals wouldn’t occur on ordinary market terms, given the government’s ability to affect their costs, markets, and future growth. Meanwhile, the administration’s existing equity deals have proceeded without clear government-wide statutory authority.
Government Should be a Neutral Party
Government ownership would blur lines that should remain clear. Namely, the government should act as a regulator, the market should allocate capital, and private firms should remain independent developers of technologies.
The federal government already regulates AI companies. It can also use procurement and contracting decisions in ways that affect a firm’s ability to compete. The recent Pentagon-Anthropic dispute shows how quickly those powers can become a fight over product design, market access, and constitutional rights. As Jennifer has written, the government labeled Anthropic a “supply-chain risk” because the company refused to change its product to meet Pentagon demands, even though the government could have just canceled the contract and chosen another vendor.
If the government also becomes a shareholder, it would have financial and political incentives tied to the success of some firms over others, possibly making it harder for smaller or newer companies to challenge government-backed incumbents and weakening the market’s competitiveness.
A regulator may hesitate to enforce rules that could reduce a government-backed company’s valuation. A procurement office may favor a company in which the government has a direct interest. Or a president may pressure companies to serve political goals while presenting that pressure as stewardship of the public’s investment.
This is the broader danger of the government-ownership model: it shifts corporate decisions toward political rather than market-driven incentives.
It is a particularly dangerous time to make such a decision around AI winners and losers. The industry is in an early and dynamic phase. As the history of innovation shows, it is not always clear who the long-standing giants will be or what all the potential applications of a general-purpose technology are. Determining winners and losers now through government investment risks putting a thumb on the scale in favor of certain companies and, more generally, disrupting private investment, as our colleague Scott Lincicome has written regarding the Intel stake.
The risk of the government picking the wrong winner at an early phase is high. In many ways, AI in 2026 is where the internet was in 1996. We can now laugh at headlines about “Will MySpace Ever Lose Its Monopoly” or how “Yahoo Won the Search Wars,” but what if the government had invested in them as the chosen champion?
While we, as a society, may occasionally be nostalgic for the Y2K internet, it was innovation and entrepreneurs who found better ways to respond to consumers’ desires, not the government.
Free Expression Concerns
Beyond that, a government interest in AI could also raise unique free expression concerns. Particularly if the government were to make the type of significant investment under Sanders’s proposals, it raises the question of how the government could potentially dictate what is and is not allowed under an AI’s algorithm. This could allow the government to potentially limit the use of such tools for criticism or counter viewpoints. It could also enable controlled or limited access to certain information. This is not because chatbots or AI algorithms have speech rights, but because the humans designing AI algorithms have free expression rights in designing the algorithm, as Jennifer and our Center for Constitutional Studies have argued. For example, how might this play out in the friction between Anthropic and the Pentagon if the government were both a significant shareholder and a customer?
There are also free expression and privacy concerns about what access to user data such actions could give the government. Could a government stake in AI companies be the new data broker loophole to access data on citizens that the government would not otherwise be able to obtain? Users have speech rights under the First Amendment, including the right to ask questions and to engage in anonymous speech. Could government control or ownership impact privacy and speech?
AI Needs a Framework, Not Federal Ownership
Rejecting government ownership doesn’t mean rejecting any AI policy framework. It means asking what kind of framework is appropriate for a general-purpose technology.
As Jennifer articulated, AI policy should follow four broad principles. Policymakers should first ask what existing law already covers. They should avoid a disruptive patchwork of state rules. They should improve AI literacy and education to address concerns about labor disruption and disinformation. And they should place safeguards on the government’s own use of data and AI to protect civil rights and liberties.
This type of light-touch approach to a general-purpose technology does not direct potential applications. It instead directs appropriate responses to specific application concerns, such as fraud, civil rights abuses, or cybersecurity. Such an approach focuses on the specific applications at the heart of the concerns without blocking innovation.
Policymakers should approach AI with the same regulatory humility they should have brought to the early internet. Policymakers shouldn’t presume that they know in 2026 where AI is headed any more than they could have known in 1996 where the internet was headed.
If the government directs the path of innovation, AI is less likely to develop in ways that meet the needs of consumers, workers, and entrepreneurs.
AI’s “Problems” Have Better Policy Solutions than AI Nationalization
Humility matters for the underlying tech, labor, and infrastructure policy debates at the heart of much of the AI debate as well.
Many AI concerns aren’t unique AI questions—they are extensions of ongoing policy debates over speech, online youth safety, privacy, cybersecurity, labor markets, and government use of technology. AI may change how those questions arise, but that doesn’t mean Washington should own the companies building the tools. If anything, as we described above, government investment could lead to more problems on key issues such as privacy and speech.
Similarly, a more complicated picture of AI and labor is emerging. AI may change work, but that’s not the same thing as an economy-wide jobs collapse. A recent paper finds that AI exposure can reduce demand for certain tasks, but AI adoption can also increase productivity and labor demand. That supports a careful response focused on adjustment, education, and AI literacy rather than a permanent government ownership regime.
The same is true of data centers. As with any infrastructure element, there are debates about the impact on resources. Electricity and water concerns can be real in particular communities, but the backlash is outrunning the evidence. Concerns about water use are frequently overstated, and evidence that data centers are broadly driving up electricity prices remains at best mixed. If electricity demand is the concern, the answer is electricity abundance, faster permitting, better siting, and more flexible grid policy. Our colleague Travis Fisher’s proposal for consumer-regulated electricity offers one possible way to let large users finance their own power needs without shifting costs onto households.
Conclusion
Policymakers often describe AI as a race with China, but there is no finish line. It’s an ongoing process of innovation, adoption, infrastructure buildout, and improvement. We are only at the beginning of understanding how AI may be applied across many industries and in our daily lives.
The US won’t strengthen its position by making leading firms more dependent on federal permission and boardroom influence. As technology and innovation expert Adam Thierer emphasizes, the US “became a technological leader because it developed policies that encouraged robust private-sector entrepreneurialism and openness to new ideas and change.”
A better AI agenda would keep barriers to entry low, enforce existing laws against concrete harms, expand energy supply, allow infrastructure to be built, and avoid policies that lock in today’s largest players. That way, AI’s benefits can be developed through experimentation and competition rather than dictated by politicians and bureaucrats in Washington.