The US president announced on Friday on his social network the forthcoming appointment of an "AI tsar" and the creation of an "AI Force", modelled on the Space Force, to oversee the industry. The precise contours remain to be defined. The choice of institutional model, however, is already telling.
We reported last week that the debate over pace had spilled into the public sphere, with a presidential intervention in favour of speed. Here is the institutional translation of it.
What the chosen model says
The Space Force is a branch of the armed forces. Choosing this model to oversee an industrial sector is not neutral: it places AI in the category of defence matters rather than that of products to be regulated.
That classification has concrete consequences. A defence matter is governed by secrecy, speed and competition between states. A regulatory matter is governed by transparency, deliberation and the protection of users. They do not use the same instruments, nor do they share the same priorities.
It is consistent with a sequence we have documented: the portal opened to three million personnel, then the mooted five-billion loan to a compute startup. The centre of gravity of US policy in this area is shifting towards defence.
What a tsar can and cannot do
A single coordinator has a real advantage: it brings together expertise scattered across numerous agencies, and gives the sector a point of contact.
It also has a structural limitation: it depends entirely on the person who appoints it. A tsar is not an independent authority. It can be replaced, its direction changes with the political majority, and it generally has no binding power of its own.
It is the exact opposite of what we described as necessary: a public oversight body capable of revising decisions, third-party assessments with proper resources, and protected whistleblowing. Those three elements presuppose a lasting institution, not a person.
Two approaches are emerging with increasing clarity. One treats AI as a matter of power, entrusted to a structure modelled on a military branch and to an appointed coordinator. The other treats it as a product subject to written rules, enforced by administrative authorities, with judicial remedies. We have covered European transparency obligations, and saw this week a company relocating to Germany over applicable law. These two models will coexist, and companies will have to contend with both.
What remains unclear
An announcement made on a social network is not a text. We do not yet know what powers this structure would have, what legal basis it would rest on, or how it would fit together with existing agencies and the work of Congress.
This vagueness is itself information. It suggests that the announcement precedes the design, which is common for presidential initiatives and rarely a good sign for their legal robustness.
What to take away
The question is not whether the United States should have a structure for AI — it clearly needs one, and the absence of federal coordination is a problem we have often noted.
The question is one of model. Treating a technology as a military matter produces institutions designed for speed and secrecy. Yet what is most lacking today, as we write in every article, is precisely the opposite: verification, transparency and duration.