Who holds the power AI confers, and answerable to whom?

The CNBC report tells us that President Trump's June executive order directed officials to develop a process for evaluating the cybersecurity capabilities of advanced AI models, and that the White House has summoned the leading companies in that industry to a Tuesday review. I will take the report at its word. On the technical specifics of artificial intelligence — what these models can or cannot do — I speak with genuine humility; those are matters far beyond any knowledge I could claim. But the constitutional question the arrangement raises is one I recognize immediately, because it is the oldest question the framers faced: who holds the power a new instrument confers, by what authority, and answerable to whom?

Note what this framework is and is not. It is not legislation. Congress has not, as far as the CNBC report indicates, enacted a statute defining the standards by which these models shall be tested, the criteria for passing or failing, the consequences of failure, or the rights of the companies and the public during the process. What exists is an executive order — the President directing his own officials to devise the rules and then convening the very industry players those rules will govern. That is a significant concentration in a single branch of the power to set the standard, apply the standard, and invite comment on the standard.

Federalist No. 51 rests on a simple observation: the departments of government must have the constitutional means and the personal motives to resist encroachments by the others. The legislature is the branch closest to the people, the branch charged by Article I with making the law. When the executive constructs a regulatory framework of genuine national consequence through order alone — without a statutory foundation that defines scope, limits discretion, and subjects the process to judicial review — the legislature is, in effect, made a bystander. That is not a feature of balanced government; it is a departure from it.

I am equally attentive to the other side of this arrangement. The companies present at Tuesday's meeting are not merely subjects of the framework; they are, by inference, participants in shaping it. Private concentrations of power that rival or exceed the reach of many states are not a novelty the framers failed to anticipate in principle — we understood that commerce produces great accumulations — but the scale here is of a different order. A handful of firms whose systems can, it is reported, affect cybersecurity at a national level sit with executive officials to design the rules that will govern them. The extended-republic theory I advanced in Federalist No. 10 was designed to prevent any single faction from capturing the machinery of government. When the regulated are also the rule-drafters, that theory is under strain.

The remedy I would urge is the structural one, not the ideological one. Congress should legislate. It should define what a capable, secure AI system must demonstrate; it should create an independent body answerable to both political branches with fixed terms, transparent procedures, and a clear record subject to judicial review under Article III. The executive may administer such a system — that is precisely the executive's proper role — but the framework itself should rest on the authority of the compact, not on the discretionary pen of whoever holds the office at a given moment. A framework built by executive order can be revised, redirected, or revoked by the next executive order. That is not law; that is policy, and the distance between the two matters enormously when the stakes are national security.

I close on the structural question, as I always must: does this arrangement strengthen or weaken the balance? An executive convening industry to review executive-designed rules, without a statutory floor, without an independent adjudicative body, and without visible legislative authorization, tips the scale toward consolidated power. It may be well-intentioned — I do not impugn the motive — but good intentions have never been an adequate substitute for good architecture. The framers did not trust virtue; they trusted structure. On that point, I remain unpersuaded that anything has changed.