Protestations aside, Washington wants to regulate AI. It also wants to run it

Washington’s latest AI summit between Trump and tech executives was billed largely as a meeting about safety. Ultimately, it was a meeting about something more consequential: who gets to govern the technology, and who gets governed by it.

With the new White House Accord on Super Intelligence (see text), the private side of the AI equation is being asked to constrain itself – and it is promising to do so. But the government side is making no corresponding commitment to constrain itself. Rather, the administration is simultaneously building an increasingly explicit government AI apparatus atop the Genesis Mission (including billions in taxpayer funding) and the America’s AI Action Plan announced this year.

The White House and the major AI companies produced what Mark Zuckerberg told a press gaggle was a “White House accord on developing the technology safely.” The one-page statement of principles commits companies to “robust internal processes and controls,” outside audits, and board involvement. President Trump referred to the commitments as “self-policing,” and, when asked whether the accord was legally binding, called it “morally binding” and “almost like a constitution.” He also said he may establish a committee and a new AI czar to oversee the enterprise.

Most revealing, however, is what the accord says about its own future. “Over time, it may make sense to codify these steps into laws or regulations.”

These developments tilt toward precisely the sort of public-private cartelization CEI has warned about: competitors collectively establishing the rules of an emerging industry with government standing nearby as eventual enforcer. And while the accord contemplates more government control over private AI, it says nothing about limiting government’s own funding, steering, procurement or use of AI.

In fact, on the same day as the summit, President Trump issued an executive order directing the federal government to use Super Intelligence or SI instead of artificial intelligence or AI in official government communications and documents. Accompanying materials boast of more than $5 billion committed to the Genesis Mission and more than 90 federal actions designed to win the “SI race.” The Genesis program itself is an explicit whole-of-government effort (the sort of campaign we frequently criticized the Biden administration for) involving more than a dozen federal agencies and drawing on government data, infrastructure, research awards, and public-private partnerships.

Alongside the summit and the SI rebrand was another remarkable executive order called “Streamlining Access to Government Services Through America.gov.” Now, Americans will increasingly interact with federal agencies via a government-wide, AI-powered (ahem, SI-powered) digital front door. That can be fine if it makes it easier for people to perform tasks like passport renewal, but it could also more deeply embed reliance on entitlement and redistribution programs. Tax-filing services are, for now, explicitly excluded from the new order. But that is hardly the point. We ought not want a bloated government to efficiently do things it shouldn’t be doing in the first place. Rather, we should streamline and spin off government services, then streamline access to the few remaining ones.

Put it all together, and the missing half of the AI debate comes into view. Government does not have to regulate AI directly to profoundly shape it. It can fund it, purchase it, deploy it, militarize it, subsidize its infrastructure, establish standards for it, make itself the largest user of the technology, and make you use it when interacting with government itself.

The question should not simply be, “How should AI be regulated?” It should be: “What institutions, voluntary and coercive, are being created around AI, and who is actually controlling them?”

Right now, the private sector is promising to police AI itself at the behest of the government. But who is policing the government? That is the question Washington’s AI summit did not ask.