A Noun Is Not an Enforcement Mechanism

Somewhere in the Department of Justice, a memo has gone out. According to Bloomberg Law, staff must now refer to artificial intelligence as "Super Intelligence" in court — not AI, not artificial intelligence, not the defendant's software, but Super Intelligence, capital S, capital I, as though the government were briefing a deity on a scheduling matter. There is no accompanying statute, no accompanying agency, and, as far as the reporting goes, no accompanying reason given. Just a word, issued from above, to be repeated, under oath, by the United States.

The report is almost more revealing for what it leaves out. We do not know who issued the instruction, whether it was written or spoken, which offices it binds, which cases prompted it, or when it took effect. In any ordinary agency, that gap would be the scandal; here it is the method. That is not a regulatory regime; that is a vibe with a routing number. Every column I have written in this space has concerned policy that does not exist; this week the government finally produced — with great confidence and no documentation — a policy consisting entirely of a preferred noun.

Three weeks ago I noted that when President Trump asked Meta, OpenAI and Microsoft to police themselves, the only genuinely binding clause was the rebranding of the technology as "super intelligence." I assumed it was a slogan. Apparently it was a directive. It has now travelled from a stage at an AI event, through the federal bureaucracy, and into court filings — which is, if you are counting, one more place than any actual AI statute has reached this year. Twelve states have written companion-chatbot rulebooks. Congress has written none. The administration has written a thesaurus.

But do not mistake this for mere branding. A word in a court filing is an argument, and this one is load-bearing. When Justice tells a judge that the thing before it is "Super Intelligence," it is not describing a product; it is describing a category of thing that ordinary law was not built for. Super intelligence is, by definition, beyond us — beyond statute, beyond precedent, beyond the reach of twelve states and their chatbot disclosure rules. The word argues for its own exemption. You do not regulate what you have already announced you cannot comprehend; you genuflect, and you defer.

The contrast writes itself. On the same day Bloomberg reported the DOJ's vocabulary rule, Geoffrey Hinton — Nobel laureate, godfather of the field, a man who has spent recent years on a one-man apology tour for the thing he helped build — proposed an FDA-style pre-market approval regime for AI, under which companies would have to convince a regulator their products were safe before release. That is a serious idea with a serious cost and a serious fight attached: a body, a budget, a standard, and a vote. The government's actual answer, this week, is to keep the noun and skip the test. Hinton wants the FDA; Washington offers the adjective. It is also, notably, the sort of proposal that requires the government to hold an opinion about the technology — which, this week, it has declined to form.

This is how a regulator dies: it begins speaking the industry's language instead of the statute's. "Drug" triggers a test, "medical device" triggers a classification, "generally recognised as safe" triggers a burden of proof. Each of those words is a doorway into a legal process. "Super Intelligence" is a doorway into awe — not a term of art but a term of advertising, now installed by memo on the government's own side of the courtroom. Former regulators have already said that the AI laws on the books are simply being ignored. We now have a government that ignores its own statutes and enforces its own adjectives.

None of this is accidental. Rebranding is the cheapest policy on the menu: no adversary, no appropriation, no committee, no lobbyist to buy off, because it does nothing. Actually pre-empting the twelve state rulebooks would take a bill and a fight; actually imposing pre-market review would take all of the above. Actually enforcing the laws already on the books would take the one thing this administration has no appetite for — an admission that the technology is ordinary enough to be governed. A word costs nothing, offends no donor, and lets the base see a government in motion. Had they spent on a statute the energy they spent on "Super Intelligence," they would have a statute — they just do not want one.

So here is the ledger for October 2026. Twelve states: rules; Congress: nothing; the frontier labs: a self-regulatory body modelled on FINRA, with no SEC behind it; the Department of Justice: a preferred term. The one durable artifact of the American AI policy era may yet prove to be a piece of rebranding, enforced in federal court by lawyers obliged to say it and unable to say why. I can tell you exactly what this administration's AI policy is: three syllables, capitalised, and not a law.