Renaming AI 'Super Intelligence' Changes No Definition in Force

The White House’s own summary of the President’s speech to the UN General Assembly on 22 September gives the line in full: “The United States totally rejects any attempt to construct a globalist scheme of control for the Artificial Intelligence being spoken of so much now — hereinafter officially called ‘Super Intelligence.’” In the room he went further. Deseret News transcribed it as: “From this point forward, all of the United States’ documents — and hopefully the world’s — will be changed to use the much more accurate term ‘super’ as opposed to artificial.” The reason he gave, per UN News: “The use of the word artificial makes intelligence fake. It is not fake.”

“Hereinafter” is a drafting word. It is what a statute says just before it defines something. So I took the sentence at its word and asked what it would have to change: which documents define artificial intelligence for the United States, and whether any of them moved.

None had moved as of this morning. The Federal Register’s full-text search returns zero documents, ever, containing the phrase “super intelligence.” The newest presidential documents in it concern H-1B visas, saltwater angling and hunting. A speech is not an instrument, so none of that is surprising. The better question is what the rename would collide with if someone did try to carry it out, because the word it borrows already has a job, and as of today a Senate bill would attach a prison term to it.

The definition everything points to

The operative definition of artificial intelligence in American law is 15 U.S.C. § 9401(3), from the National Artificial Intelligence Initiative Act, enacted on 1 January 2021 as part of that year’s defense authorization:

a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments.

It is deliberately plain. It says nothing about how clever the system is. A spam filter qualifies, and so does whatever the frontier labs ship next year.

This administration has used that definition consistently. I pulled every executive order signed since 20 January 2025 that the Federal Register’s full-text search matches on “artificial intelligence”, and read each one. There are 26. Twenty-two use the term without defining it, including the two most consequential AI orders of the past year: EO 14365 on pre-empting state AI laws and EO 14409 on frontier-model security. The other four do define it, and all four use the same sentence: “‘artificial intelligence’ or ‘AI’ has the meaning set forth in 15 U.S.C. 9401(3).” That includes EO 14179, the administration’s first AI order, and, less predictably, an order on clean coal, which needed the definition to say which data centers it meant.

How 26 executive orders since January 2025 define artificial intelligence Sankey diagram. Of 26 executive orders signed between 20 January 2025 and 23 September 2026 whose Federal Register text contains the phrase artificial intelligence, 22 use the term without defining it, including EO 14365 and EO 14409. The remaining 4, EO 14179, 14261, 14277 and 14355, define it, and all four do so by pointing to 15 U.S.C. 9401(3). None defines it any other way. EXECUTIVE ORDERS · HOW THEY DEFINE “ARTIFICIAL INTELLIGENCE” · 2025–2026 22 orders use the term without defining it — Federal Register, read 23 Sep 2026 4 orders define it, all as 15 U.S.C. 9401(3) — Federal Register, read 23 Sep 2026 26 executive orders — Federal Register, 20 Jan 2025 to 23 Sep 2026 22 — no definition 4 — defined by reference to 15 U.S.C. 9401(3) 26 executive orders signed since 20 Jan 2025 that use the phrase “artificial intelligence” 22 use the term, define nothing e.g. EO 14365, EO 14409 4 define it, all as 15 U.S.C. 9401(3) EOs 14179 · 14261 · 14277 · 14355 0 define it any other way Federal Register full-text search, executive orders only; each order's text read 23 Sep 2026. Band height ∝ number of orders.
Every executive order of this administration that defines artificial intelligence does it by pointing at the 2021 statute. None writes a definition of its own, so changing what the term means runs through the statute.

The executive branch has its own agencies, too. On 7 July the Federal Trade Commission published a policy statement on AI systems that suppress accuracy. It footnotes § 9401(3) for what AI does. It is also the only document in the Federal Register’s full-text search, in any year, that uses the word “superintelligence.” It uses it exactly once, in a list of things AI might be: “modern large language models, AI applications applying large-language models to particular purposes, or a not-yet-developed superintelligence.” Eleven weeks before the speech, an agency of the same administration was using the word for the one kind of AI that does not exist yet.

What the word already means

That matches how the word is usually defined. Nick Bostrom’s 1998 definition is “an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills.” It is a comparison with people, and a steep one. The statutory definition of AI makes no comparison with people at all.

The comparison now has a bill trying to write it into law. On 23 September Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act. When I wrote about it ten days ago it existed only as a press release and a one-page summary. Now there is bill text, and its definition is tighter than the summary’s. “Match or exceed” has become “exceeds,” and “a broad range of domains” has become “most domains.”

A cropped page of Senate bill text, lines 3 to 17 of page 5, defining 'artificial superintelligence' as an artificial intelligence system that exhibits, or can easily be modified to exhibit, either of two capabilities: (A) exceeding human cognitive performance and capabilities across most domains or tasks, including decision making, learning and adaptive behavior; or (B) sufficient capabilities to plan and execute the destruction or disempowerment of humanity, including by overthrowing or undermining the Federal Government.
The definition that section 9 of the bill prohibits anyone from developing, deploying, possessing or funding. Prong A is a comparison with human performance; the words "or can easily be modified to exhibit" reach systems that do not meet it yet. Image: Office of Senator Bernie Sanders, "Ban Artificial Superintelligence Act," bill text, page 5, lines 3–17. US government work, public domain. Cropped.

Section 9 of the bill says no person may “develop, deploy (either internally or externally), acquire, possess, fund, import, or transfer” artificial superintelligence. Section 13 then provides that a policymaking employee of a chartered AI company, or a “rogue actor” outside one, who recklessly violates it “shall be fined under title 18, United States Code, and imprisoned for not more than 20 years.” The bill has only been introduced. What matters here is the direction it pushes the word: towards a narrow, high threshold that a prosecutor would have to show was crossed.

Put the two usages from 22 and 23 September next to each other and the word has been stretched in opposite directions inside a day. If US documents did start calling all AI “Super Intelligence”, every recommendation engine would carry, in official prose, the name that a pending bill would reserve, on pain of prison, for systems better than people at most things. The bill never defines “artificial intelligence system” itself. Under a literal rename, anyone reading it would first have to work out which “superintelligence” it meant.

How much of the capability range each text covers Schematic range chart with an ordinal horizontal axis running from narrow single-task software on the left to beyond the best humans on the right. 15 U.S.C. 9401(3) and EU AI Act Article 3(1) cover the whole range, with no capability floor or ceiling. The 22 September 2026 UN speech applies the name Super Intelligence to that same whole range. The Ban Artificial Superintelligence Act's prong A covers only the region beyond exceeding humans across most domains, with a dashed extension of undefined length for systems that can easily be modified to get there. Bostrom's 1998 definition is a point at the far right: much smarter than the best humans in practically every field. WHICH SYSTEMS EACH TEXT REACHES · SCHEMATIC, NOT TO SCALE exceeds humans across most domains 15 U.S.C. 9401(3) EU AI Act, Art. 3(1) UN speech, 22 Sep 2026 Ban Artificial Superintelligence Act Bostrom, 1998 in force since 1 Jan 2021 Regulation (EU) 2024/1689 “Super Intelligence” for all AI prong A, introduced 23 Sep 2026 “much smarter than the best human brains” Whole range: no capability floor or ceiling — 15 U.S.C. 9401(3) Whole range: no capability floor or ceiling — EU AI Act Art. 3(1) Whole range: the name applied to all AI — UN General Assembly, 22 Sep 2026 Beyond exceeding humans across most domains — bill text, sec. 3(2)(A) Much smarter than the best human brains in practically every field — Bostrom, 1998 “can easily be modified to” narrow, single-task software beyond the best humans Axis is ordinal, not measured. The dashed extension's length is undefined in the bill. Prong B (capability to disempower humanity) is not a point on this axis and is not drawn.
The statute and the EU regulation cover every capability level. The bill and the older literature use "superintelligence" for the top end only. The speech puts that top-end name on the whole range.

The word that does the work

The speech’s complaint was about the adjective: “artificial” sounds fake. The legal work in § 9401(3) is done by a different word, “human-defined.” The US definition covers systems pursuing objectives that people set. That qualifier is not universal. The EU AI Act’s Article 3(1) says “explicit or implicit objectives” instead. Luciano Floridi noted in 2023 that the OECD’s revised definition had dropped “human-defined” too, and argued that it was a mistake.

Whether a system’s goals are set by people is a large part of what the superintelligence literature worries about. So the American definition’s most distinctive clause is the one a superintelligence in Bostrom’s sense might stop satisfying. Renaming the category after that case leaves the clause untouched, and blurs the difference the clause is there to mark.

Recent governance research has mostly moved the other way, away from a single noun and towards graded properties. Atoosa Kasirzadeh and Iason Gabriel’s agentic-profiles framework, published in Nature in August, rates systems separately on autonomy, efficacy, goal complexity and generality, each on its own scale. I leaned on it a few days ago for shutdown policy. Its premise is that one word can’t tell a regulator what it needs to know about a system. The speech tries to do it with a word that tells the regulator even less.

What the rename was attached to

The sentence’s substance came before the rename: the rejection of “any attempt to construct a globalist scheme of control.” That had a specific target. The day before, 20 countries and the European Union, including Germany, Canada, Australia, Singapore and the UAE, had issued a joint statement calling for AI to stay under “human direction, oversight and control” and proposing an international institution to “set standards, enable verification, and convene states when capability thresholds are crossed.”

“Capability thresholds” is what the word superintelligence was supposed to name, and what the bill’s prong A tries to write down. It is the line at which a system becomes a different kind of problem. Using the word for every system removes that line from the vocabulary. The countries asking for thresholds and the President rejecting them now use different dictionaries, which will make any conversation between them harder than it already was.

The name “artificial intelligence” has an unusually traceable birth. It appears in the title and the first sentence of a proposal dated 31 August 1955, from John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon: “We propose that a 2 month, 10 man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College.” It was a grant request with no legal force. It took until 2021, 65 years later, for Congress to give the phrase a statutory definition. That definition, not the name, is what carries weight in an executive order.

A future order could try to change the label. If it followed the four orders above, it would still define its new term as “has the meaning set forth in 15 U.S.C. 9401(3)”, and that meaning would still include the spam filter. Changing the meaning itself takes an act of Congress. The bill that does define the word uses it to mark the systems it wants to prohibit.

References

  1. The White House (2026). President Trump at the United Nations: “While Others Have Talked, I Have Acted”. 22 September 2026. Source of the “hereinafter officially called ‘Super Intelligence’” sentence.
  2. Deseret News (2026). Trump delivers remarks at U.N. General Assembly amid Iran war, teases new ‘super intelligence’. 22 September 2026. Transcription of the “all of the United States’ documents” passage; Gizmodo’s transcription differs slightly in wording.
  3. UN News (2026). Trump defends military action against Iran, Venezuela, and calls for rebrand of AI. 22 September 2026.
  4. Federal Register. Document search API, queried 23 September 2026: zero documents containing “super intelligence”; 26 executive orders signed since 20 January 2025 containing “artificial intelligence”, each read in full. The counts are my own.
  5. 15 U.S.C. § 9401(3). Definitions. National Artificial Intelligence Initiative Act of 2020, Pub. L. 116–283, div. E, 1 January 2021.
  6. Executive Order 14179 (2025). Removing Barriers to American Leadership in Artificial Intelligence. Signed 23 January 2025.
  7. Executive Order 14261 (2025). Reinvigorating America’s Beautiful Clean Coal Industry and Amending Executive Order 14241. Signed 8 April 2025.
  8. Executive Order 14277 (2025). Advancing Artificial Intelligence Education for American Youth. Signed 23 April 2025.
  9. Executive Order 14355 (2025). Unlocking Cures for Pediatric Cancer With Artificial Intelligence. Signed 30 September 2025.
  10. Executive Order 14365 (2025). Ensuring a National Policy Framework for Artificial Intelligence. Signed 11 December 2025.
  11. Executive Order 14409 (2026). Promoting Advanced Artificial Intelligence Innovation and Security. Signed 2 June 2026.
  12. Federal Trade Commission (2026). Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems. Federal Register doc. 2026-13628, 7 July 2026.
  13. Bostrom, N. (1998). How Long Before Superintelligence? Historical background.
  14. Office of Senator Bernie Sanders (2026). Sanders, Casar Introduce Legislation to Create New Federal Agency to Ban Artificial Superintelligence, Pause Advanced AI Development. 23 September 2026.
  15. Office of Senator Bernie Sanders (2026). Ban Artificial Superintelligence Act, bill text. Sections 3(2), 9(a) and 13(b).
  16. European Union (2024). Regulation (EU) 2024/1689 (AI Act), Article 3. Background.
  17. Floridi, L. (2023). On the Brussels-Washington Consensus About the Legal Definition of Artificial Intelligence. Philosophy & Technology 36(4). Background.
  18. Kasirzadeh, A. and Gabriel, I. (2026). Agentic profiles for effective AI governance. Nature 656(8127), 320–328.
  19. Power, J. (2026). 20 countries propose global oversight body to manage AI dangers. Al Jazeera, 22 September 2026.
  20. McCarthy, J., Minsky, M. L., Rochester, N. and Shannon, C. E. (1955). A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence. 31 August 1955. Historical background.