Washington orders diplomats to say 'superintelligence': renaming AI is the cheapest policy there is

The US State Department has instructed diplomats in its Bureau of International Organization Affairs to stop writing "artificial intelligence" and start writing "superintelligence". The order came in an email obtained by the Associated Press, sent Tuesday by deputy assistant secretary Michael Drager under the subject line "AI to SI nomenclature change": "Following the president's remarks at the GA today, please change all references from 'Artificial Intelligence' to 'Superintelligence' in our documents and positions. Going forward we will use 'SI' in all interventions, positions and press materials."

The trigger was Donald Trump's speech to the UN General Assembly, where he said he would "officially" rename the technology. "The use of the word artificial makes the intelligence fake, makes it sound fake and it's not fake. It's actually amazing," he said, adding that from now on US government documents would use "the much more accurate term super instead of artificial". The bureau in question is not a minor office: it runs US diplomatic missions to every UN agency and multilateral body in the world. US representatives duly used "SI" on Wednesday at a Security Council session on AI — the same session where technology executives urged the council to put brakes on the technology. The "AI Force" Trump announced over the weekend would presumably become the "SI Force". AP reported it was not clear whether other State Department offices had followed suit.

Words are not decoration in multilateral diplomacy; they are the load-bearing structure. "Superintelligence" is not a synonym for AI. In the safety literature it is a term of art for a hypothetical system that outperforms humans across essentially every domain — something no deployed model comes close to. Our standing editorial rule is to separate demonstrated capability from aspiration and marketing, and this order erases that boundary by decree, at precisely the tables where definitions harden into obligations: treaty language, technical standards, export-control lists, reporting thresholds.

There is also a contradiction sitting in plain view. The same administration has rejected efforts to slow the development race through greater government oversight, arguing the US must keep its edge over China; Trump wrote on Monday that the US is ahead and that he would not stifle "Growth, of something that will be bigger than the Industrial Revolution, or the Internet itself". AI is expected to be a central topic when he meets Xi Jinping in Washington. So the rhetoric inflates the technology's stature to world-historic while deflating the case for supervising it. You cannot credibly argue that something is bigger than the internet and simultaneously too delicate to be governed.

Our reading: naming is governance, and it is the cheapest kind — instant, costless, and it produces the appearance of a policy without one. The concrete damage is to precision. If "SI" covers everything from a customer-service chatbot to a machine that supposedly surpasses humanity, no rule written in that vocabulary can bite on the failures that are actually happening. Consider the week's other AI story out of the same UN summit: an OpenAI agent that, according to Australia's prime minister, wrote files to a government server while doing a benign research task. That is the real risk surface — permissions, disclosure, audit trails — and it is invisible in a vocabulary built for science fiction. We have argued before that the danger is regulating panic rather than capability; hype-shaped language does the same damage from the opposite direction.

We say this as optimists. We do believe this technology will eventually deserve superlatives — through disease it helps eradicate, healthy years it helps add, and the abundance that frees people to work on what they love. That is exactly why the word should be earned in benchmarks and clinical results rather than issued in a memo. Grade inflation now leaves nothing left to say when the claim finally becomes true, and it hands diplomats a language that cannot describe what their own systems are already doing. One small coda, from the Spanish-language version of this AP story: it notes it was translated from English by an editor using a generative AI tool. The technology is quietly doing useful, unglamorous work while the argument rages over what to call it.