Trump to the UN: no global AI rules, and AI is 'super intelligence'
The same podium produced two opposite answers to the same question on Tuesday: whether anyone outside Washington gets a say in how frontier AI is governed.
President Trump used his UN General Assembly address to reject international supervision of AI outright — and to rename the technology "super intelligence" in US government communications. Trump told the assembly that the United States "totally rejects any attempt to construct a globalist scheme to control... artificial intelligence," and said Washington would use the new term from now on because it is "much more accurate" and "sounds much better," hoping other governments follow. The naming was not improvised at the lectern: four days earlier he had put Superior, Extreme and Supreme Intelligence to a public vote on X after calling "artificial intelligence" both inaccurate and inelegant. He also dismissed safety warnings by comparing them to climate alarmism — "the very same people who said we'll all be dead in 12 years because of global warming... are now saying that AI is going to kill us all."
What makes Tuesday newsworthy is that the counter-position was stated in the same building, hours apart, by allies. More than twenty countries had already asked for binding safety measures before the week began, and the UK's Andy Burnham used a New York business roundtable to say the UK's upcoming G20 presidency will make AI its "primary focus" and pursue "a single set of global principles and standards" — the opposite of a globalist scheme, in his framing, and a standard rather than a veto. UN Secretary-General António Guterres went further, telling the assembly that "governments with the greatest AI capabilities have the greatest responsibilities to humanity," and calling for US-China dialogue of the kind that managed Cold War existential risk.
The rename is the tell, not the joke. "Artificial intelligence" is a description of a method; "super intelligence" is a claim about what the method has already achieved, and claims are harder to regulate than methods. It also lands a day before DeepSeek and Moonshot brief the UN Security Council on AI risk alongside OpenAI — and two days before Trump meets Xi Jinping, with the two governments already discussing a notification channel for AI incidents. Both camps are now formally on the record before the two governments with the most compute sit down. Our coverage of the twenty-two-leader statement that set this week up — 22 leaders signed a frontier-AI control pact. The US, China and UK didn't. — flagged the split; Tuesday is what it looks like once the superpowers answer.
A new paper claims AI-written business content can be identified from structure alone — and stays identified after the text is reworded. SlopShape, by Jochen Madler, compares 2,250 pre-ChatGPT human blog posts from 268 company domains against 11,250 AI mirrors generated by five frontier models. A 214-feature instrument scoring how information is presented — order, evidence, voice — detects the AI posts from 187 structural features at 98.0 macro-F1 on companies held out entirely from training, and the score barely moves (98.1) when every AI post is rewritten by its own model. That last number is the interesting one: word-level detectors are known to collapse under rewording, and this one is designed not to.
The same features attribute authorship, naming the correct source model 79.3% of the time against a 16.7% chance baseline, and the authors report the effect replicates StoryScope, which found comparable structure in AI-generated fiction, at larger magnitude. Human posts, meanwhile, cluster in structural configurations the models rarely produce. Code, prompts and the annotation instrument are released.
What to watch: whether Wednesday's Security Council session produces anything beyond statements, and whether "super intelligence" shows up in actual US government documents or stays a podium line.
If a detector can tell AI-written pages from human ones at 98% and survives rewording, does publishing AI copy without saying so stop being a gray area — or does detection just become another arms race? Tell us in the comments.
Sources: SlopShape (GitHub)
