Executive Order 14434 tells federal agencies to write SI where they wrote AI, yet it defines SI as exactly what US law already calls AI. Meanwhile researchers and senators use the same words for systems that do not exist, and an older school uses SI for machines that are real rather than imitated. The 28 November definition decides which meaning gets legal force.
Table of Contents
On 29 September 2026 President Donald Trump signed Executive Order 14434, which directs every executive department and agency to write “Super Intelligence” and “SI” in place of “artificial intelligence” and “AI”. The order then defines the new term as the technologies that section 9401(3) of title 15 of the US Code already calls artificial intelligence. The vocabulary changed that day. The law did not.
That gap is the story, because “superintelligence” was never an empty word waiting for a government to claim it. The philosopher Nick Bostrom defined it in the 1990s as an intellect far ahead of the best human minds in practically every field, and the companies building frontier models still describe it as a stage no system has reached. A bill introduced in the House five days before the order would ban a related category outright. A third tradition, older than either, has used SI as the abbreviation for synthetic intelligence, a name proposed so that machine intelligence would count as genuine rather than imitated.
Three claims now sit on the same two letters, and only one is attached to a deadline. By 28 November 2026 the White House science adviser must deliver proposed legislative language that could keep SI as a synonym for AI or turn it into a capability threshold.
An order that renames the vocabulary and keeps the statute
Executive Order 14434, titled Inaugurating the Era of Super Intelligence, was published in the Federal Register on 2 October 2026. Its first section argues that today’s frontier systems far exceed what was envisioned when the term artificial intelligence came into use, and that they do more than imitate or automate discrete parts of human intelligence. On that reasoning the administration concludes that federal terminology should reflect the capability, and that “Super Intelligence” does so better. The president made the same case in the room on 29 September, telling the executives beside him, “It’s not artificial. We all agree on that.”
The operative text is shorter than the rationale. Agencies must use the new terms, to the maximum extent the law permits, in official correspondence, public communications, websites, reports, policy documents and other non-statutory documents. The executive branch, the order says, “will not acknowledge” the old terms in any applicable setting. Regulations, contracts, grants and other historical documents issued earlier need not be altered.
Section 3 is where the order stops short of its own rhetoric. For purposes of the order, SI means the technologies and systems within the statutory definition of artificial intelligence in 15 U.S.C. 9401(3), which describes a machine-based system that, for human-defined objectives, makes predictions, recommendations or decisions influencing real or virtual environments. That definition governs until a later presidential action or an Act of Congress supersedes it. The law firm Arnold & Porter reads the result plainly: the order changes no statute, imposes no obligation on private parties and carries no penalty or enforcement provision.
Agencies are already complying. The National Institute of Standards and Technology has posted a notice that it is updating its communications, according to the same analysis, and the White House fact sheet already calls the policy framework unveiled in March 2026 the National Super Intelligence Policy Framework. The fact sheet’s argument is that the older name implies simple imitation of human intelligence, a point it makes while citing more than $5 billion announced for the Genesis Mission and the July 2025 action plan.
A 1955 grant request produced the word artificial
The order’s history is accurate as far as it goes. In 1955 John McCarthy of Dartmouth College, Marvin Minsky of Harvard, Nathaniel Rochester of IBM and Claude Shannon of Bell Laboratories proposed a summer study of “artificial intelligence”, and McCarthy asked the Rockefeller Foundation to fund it. The workshop ran in 1956, and the proposal is credited with introducing the term. All four proposers worked at American institutions, which supports the order’s claim that American scientists named the field.
The reason for the name is less flattering to the order’s argument. Early-1950s researchers used cybernetics, automata theory and complex information processing for overlapping work, and McCarthy picked artificial partly for neutrality. It avoided a narrow focus on automata theory and the analog feedback emphasis of cybernetics, and it spared him from having to accept the assertive Norbert Wiener as the field’s guru, according to the account of Nils Nilsson’s history summarised on Wikipedia. The word was chosen for neutrality among rival schools, not as a verdict that machine intelligence was fake.
The proposal’s own founding conjecture does lean toward the reading the administration rejects. It assumed that every aspect of learning or any other feature of intelligence could be described precisely enough for a machine to simulate it. Simulation was the programme. Seventy years of results have carried the field from simulating narrow tasks to general-purpose systems that produce text and images, and the order is right that the founding vocabulary was written for a smaller ambition. Whether the fix is a new word or a definition that says what the old word now covers is the question the next sections test.
Bostrom’s superintelligence describes a machine that does not yet exist
Bostrom restated his definition in a 2003 paper as any intellect that vastly outperforms the best human brains in practically every field, including scientific creativity, general wisdom and social skills. He drew the boundary deliberately. In that paper the chess program Deep Blue does not qualify because it is clever in one narrow domain, and a corporation or the scientific community does not qualify because neither is an integrated intellect. The category is general, and it sits above human level.
Industry usage follows the same logic. PolitiFact’s survey of company statements notes that Google DeepMind authors proposed in 2023 a ladder of AGI levels whose top rung, artificial superintelligence, means a general system that outperforms every human, and that the rungs above level one, where they placed ChatGPT, were not yet achieved. Sam Altman wrote in January 2025 that OpenAI was turning its aim beyond AGI toward superintelligence in the true sense. Meta created Meta Superintelligence Labs in June 2025, and Microsoft AI described a superintelligence team pursuing a humanist version in November 2025. Each used the word for a goal, not for a product on sale.
The company statements after 29 September show the strain. OpenAI’s chief global affairs officer, Chris Lehane, called the question of whether the technology has reached superintelligence subjective in a 2 October CNN interview. Elon Musk said he would rename SpaceXAI as SpaceXSI, and Nvidia chief executive Jensen Huang called data centres superintelligence factories at the White House event. Otherwise, PolitiFact found on 7 October, the public materials of OpenAI, Google DeepMind and Anthropic showed little to no change in how they refer to AI.
The scientific record cuts against calling today’s systems superintelligent. PolitiFact cites the 2026 International AI Safety Report, which describes general-purpose AI as jagged, strong at some tasks and weak at others. A system that is jagged across tasks does not meet a definition that requires superiority in practically every field.
Three vocabularies and one statute
Put the claims side by side and the mismatch becomes mechanical. The executive branch now uses SI for everything the statute calls AI. Bostrom and the frontier labs use it for a capability nobody claims to have. The Ban Artificial Superintelligence Act, filed in the House as H.R. 10538 on 24 September, defines artificial superintelligence as a system that matches or exceeds human cognitive performance across a broad range of domains or tasks, or that can plan the disempowerment of humanity. Its sponsors, Sen. Bernie Sanders and Rep. Greg Casar, would back the prohibition with up to 20 years in prison for individuals and what they call a corporate death penalty for entities.
Meanings attached to the letters SI and its neighbours
| Claimant | Term | What it covers | Does the capability exist |
|---|---|---|---|
| Executive Order 14434 | Super Intelligence (SI) | Every system in the 15 U.S.C. 9401(3) definition of AI | Yes, by definition |
| Bostrom, 2003 | superintelligence | An intellect vastly above the best human brains in practically every field | No |
| H.R. 10538 | artificial superintelligence | Human-level or above across broad domains, or able to plan humanity’s disempowerment | Prohibited if built |
| Haugeland, Bratton | synthetic intelligence (SI) | Machine intelligence that is genuine rather than imitated | Disputed |
| EU AI Act Article 3(1) | AI system | Machine-based system that infers how to generate outputs | Yes, applies since 2 February 2025 |
The table lists each claimant’s term, the threshold it sets and whether the capability exists. Only the first row governs federal documents in the United States, and only the last row governs EU law.
The statutory breadth is what makes the order’s definition awkward. The European Commission’s guidelines show how wide a comparable definition runs, listing spam detection, image classification and fraud detection as examples of AI systems built with supervised learning. The US text requires only a machine-based system that makes predictions, recommendations or decisions, so by this analysis the same tools are SI under the order. Nothing in the order’s rationale suggests that a spam filter is superintelligent, yet its definition clause includes one.
Incentives explain why the word is contested. The administration says the United States leads the world and must stay ahead of China, and on 30 September the president argued that if America stops, someone else will build superintelligence and win. Senators who want a pause argue the opposite case. A name that sounds like achievement supports the first position, although the historian Thomas Haigh, quoted by PolitiFact, doubts that it reassures a public he describes as more fearful than excited. The same two words serve opposite policy positions, which is why a definition rather than a label has to settle the dispute.
Synthetic intelligence, the rival name from 1985
The letters SI had a prior owner. John Haugeland’s 1985 book Artificial Intelligence: The Very Idea proposed an analogy between simulated and synthetic diamonds. Only the synthetic diamond is truly a diamond, and a synthetic intelligence would likewise be human-made without being a simulation. Haugeland used the label for the research of his day, which he called good old fashioned artificial intelligence, according to the Wikipedia summary of his argument.
Benjamin Bratton revived the point in an essay reproduced on the P2P Foundation wiki, where he describes revisiting with researchers at the Strelka Institute the economist Herbert Simon’s older distinction between artificial and synthetic. A plastic diamond only resembles the original, while a laboratory-grown diamond is the original at molecular level. In Bratton’s words, “We should be pursuing synthetic intelligence, not artificial intelligence.” He adds a second sense, the synthesis of human and machine intelligence, and cites the 2016 AlphaGo match against Lee Sedol, in which a machine’s unexpected move prompted an equally unexpected human reply.
Philosophers have fought over the underlying question without settling it. Russell and Norvig ask whether machines can think in the way aircraft fly or in the way submarines do not swim. Drew McDermott argues the flying analogy fits, John Searle holds that a simulated mind, like a simulated rainstorm, soaks no one, and Daniel Dennett calls much of the dispute semantic, observing that a perfect copy of a Chateau Latour is a fake while any vodka is real whoever made it. Russell and Norvig conclude that for artificial minds no convention exists to say whether behaviour or pedigree decides.
The synthetic argument and the executive order overlap in one claim and diverge in another. Both reject the idea that machine intelligence is a mere imitation. Synthetic intelligence argues about kind, genuine rather than faked, while Super Intelligence argues about degree, above rather than at human level. The two words answer different questions, so adopting one does not settle the other. An SI abbreviation in a document can therefore announce a philosophical claim about genuineness or a claim about magnitude, and the reader cannot tell which without a definition.
Two vocabularies in contracts, statutes and Brussels
The IAPP, a privacy and AI governance association, frames the practical issue as managing two vocabularies. Its analysis, written by Dykema attorney Jennifer Dickey, notes that organisations will meet SI in federal agency materials while statutes, state laws, contracts, vendor questionnaires and governance frameworks keep using AI, and that the immediate change is linguistic rather than a rewrite of US AI law. Because the order says the executive branch will not acknowledge the old terms, Arnold & Porter advises companies to assess the language they use in proposals, filings and replies to agency requests, while still using the old terms when they cite existing statutes, regulations, contracts or grants.
The European Union is untouched. Article 3(1) of the AI Act defines an AI system as a machine-based system designed to operate with varying levels of autonomy that infers how to generate outputs such as predictions, content, recommendations or decisions. The definition has applied since 2 February 2025, and the Commission’s July 2025 guidelines, which are non-binding, leave authoritative interpretation to the Court of Justice. A company selling in both markets therefore holds two legal definitions that overlap but are not identical. The EU text lists content as an output and requires inference, while the US text names predictions, recommendations and decisions for human-defined objectives.
On the day of the order, Trump and the leaders of Google, Anthropic, Meta, OpenAI, SpaceXAI and Nvidia signed the White House Accord on Super Intelligence, a one-page pledge. It sets four layers of controls: internal monitoring of model capabilities, an internal team to run them, an independent external evaluator and a board committee. Arnold & Porter stresses that the accord is voluntary and that it says every company training frontier models should adopt the layers and that codification may make sense over time. Voluntary commitments of this kind can still shape procurement terms and standards of care for companies that never signed. The accord carries the new name in its title, although the order itself does not reference it.
Limits of the evidence on the rename’s effect
The order is thin on enforcement. Arnold & Porter found no penalty in it, and on 8 October the president posted on Truth Social that the White House considers anyone who says artificial intelligence instead of Super Intelligence “THE ENEMY”, without naming any consequence. A threat of that kind has no legal mechanism behind it in the order, and agencies are the only parties the order binds.
The administration’s case deserves its strongest form. Governments define terms for their own documents all the time, the order protects continuity by tying SI to the statute, and the capability language reflects real progress in systems that handle text and images. The jaggedness reported in the International AI Safety Report does not deny those capabilities. Critics make a different case. Baobao Zhang of Syracuse University told PolitiFact that definitions matter for any attempt to regulate this fast-moving technology and that systems falling outside a new definition could still cause catastrophic harm.
Several things remain unknown. No study of how the public reacts to the term was found in the sources reviewed. The poll PolitiFact cites, from Data for Progress in September 2026, found 68 percent support for a bill requiring a slowdown and a ban, which measures a position on a bill and not a reaction to a name. The order does not say whether the 28 November proposal will be published, and no text of it had surfaced when this analysis was prepared. Whether Congress will take up a new definition, and whether other governments will copy the vocabulary, cannot be inferred from anything issued so far. The available sources cannot establish whether the new word will change any decision, as opposed to any document.
Drafting rules for contracts, compliance files and published pages
Each reader group faces a different decision, and the evidence supports a different rule for each. Federal contractors and grant applicants should mirror the vocabulary of the solicitation in front of them, because agencies are instructed to avoid the old term, and should keep the statutory name when they cite a statute, regulation or existing award. The cost of mixing the two is a proposal that reads as out of step, and the cost of over-correcting is a citation that no longer matches the document it points to.
Governance and privacy leads should maintain a one-page crosswalk with three rows: SI as the order defines it, AI as 15 U.S.C. 9401(3) defines it, and AI system as Article 3(1) of the EU AI Act defines it. Each row needs its citation and an as-of date, and the file should be reopened the day the White House proposal is public. A federal document that never mentions AI may be using SI instead, so a keyword search that looks only for the old term will miss it. That is an inference from the order, not a finding from any audit.
Publishers and search editors face an ambiguity problem. SI has at least two established expansions, super intelligence in the order and synthetic intelligence in the philosophical literature, and a page that uses the letters without defining them leaves readers and answer engines to guess. Defining each use at first mention, in a sentence that stands on its own, removes the guess. Headlines that say the president replaced AI also overstate the change, since the statute, the EU definition and most company usage are unchanged.
Legislative drafters hold the most influence. A bill that uses superintelligence should state the threshold in its own text, as the Sanders and Casar bill does, and should say whether statutes that already use the word AI are affected. A statute that borrows the executive vocabulary without defining it inherits the mismatch between a label and a threshold. That drafting risk is concrete: the same word is currently attached to a definition that covers spam filters and to a prohibition that carries prison terms.
The 28 November proposal decides whether SI is a label or a threshold
The order gives the Assistant to the President for Science and Technology 60 days, which ends on 28 November 2026, to submit proposed legislative language for a federal definition of Super Intelligence. The proposal must assess whether the definition should modify, expand on or supersede the existing statutory definition of artificial intelligence, list conforming amendments to statutes that use the old term, and recommend any further executive action. It goes to the President, not to Congress, and only Congress can change the statute, as Arnold & Porter notes.
Three outcomes are possible, and each can be tested against the documents. In the first, the proposal keeps the tie to 9401(3), SI stays a synonym and the order remains a vocabulary rule. In the second, the proposal writes a capability threshold close to Bostrom’s or to the Sanders and Casar text, which would leave most existing systems outside the word SI and require amendments across every statute that currently says AI. In the third, the proposal widens the definition to cover frontier systems beyond the current text while the old statutory AI remains in force beside it.
This is an editorial forecast. The first outcome is the most likely state through the end of 2026, because the order already ties SI to the statute until superseded, because Congress controls any amendment, and because no proposal text was public as of 7 November. Until the proposal is published, SI in a federal document most likely means AI, and SI in a research paper or a bill most likely means a threshold.
Four developments would change that judgment: publication of the proposal with capability language, a bill that adopts the executive vocabulary, an agency solicitation that defines SI by capability, or movement to codify the accord. The reader’s rule is the same in every case, and it is specific: read the definition before the letters.
Reader questions about SI, AI and Executive Order 14434
In US federal documents SI now stands for Super Intelligence under Executive Order 14434. In philosophy and research writing it has long abbreviated synthetic intelligence, a term associated with John Haugeland and Benjamin Bratton. A document that uses SI without a definition can mean either.
Only inside the executive branch’s non-statutory documents. The order directs agencies to use Super Intelligence and SI to the maximum extent permitted by law, and it leaves statutes, earlier regulations, contracts and grants as they are. Congress has not changed the statutory definition.
It defines the term as the technologies and systems covered by the definition of artificial intelligence in 15 U.S.C. 9401(3): a machine-based system that, for human-defined objectives, makes predictions, recommendations or decisions influencing real or virtual environments. That definition applies until a later presidential action or an Act of Congress supersedes it.
The order gives the Assistant to the President for Science and Technology 60 days from 29 September 2026, which ends on 28 November 2026. The proposal goes to the President as draft legislative language, and the order does not say whether it will be published.
No. AGI usually means AI at roughly human level across tasks, while superintelligence, in Bostrom’s definition, means an intellect that vastly outperforms the best human brains in practically every field. PolitiFact notes that superintelligence goes beyond AGI.
PolitiFact reports that the industry generally agrees no system has achieved it. OpenAI’s Chris Lehane called the question subjective in an October 2026 interview, and the 2026 International AI Safety Report describes general-purpose AI as jagged, strong at some tasks and weak at others.
No. Arnold & Porter’s analysis finds that the order imposes no obligations on private parties and contains no penalty or enforcement provision. Companies that sell to federal agencies may still need to adjust the language of proposals and filings.
It is a proposed alternative to artificial intelligence that treats machine intelligence as genuine rather than imitated. John Haugeland compared it to a synthetic diamond, which is truly a diamond, and Benjamin Bratton argues for synthetic over artificial intelligence on the same logic.
No. Article 3(1) of the AI Act defines an AI system as a machine-based system designed to operate with varying levels of autonomy that infers how to generate outputs. The definition has applied since 2 February 2025, and the US order does not change it.
According to Arnold & Porter, the leaders of Google, Anthropic, Meta, OpenAI, SpaceXAI and Nvidia signed it with the president on 29 September 2026. The accord is voluntary and sets four layers of controls and audits for frontier models.
Author: Jan Bielik CEO & Founder of Webiano Digital & Marketing Agency
This article is an original analysis supported by the sources cited below
Inaugurating the Era of Super Intelligence (Executive Order 14434) The full order text: the purpose clause, the directive to use Super Intelligence and SI in non-statutory documents, the definition tied to 15 U.S.C. 9401(3), the exemption for historical documents and the 60-day deadline for a proposed legal definition.
Fact Sheet: President Donald J. Trump Inaugurates the Era of Super Intelligence The administration’s own statement of its rationale, including the claim that the old name implies imitation, the Genesis Mission funding figure and the renamed National Super Intelligence Policy Framework.
15 U.S. Code § 9401 – Definitions The statutory definition of artificial intelligence that Executive Order 14434 adopts as the meaning of Super Intelligence, as published by the Cornell Legal Information Institute.
Commission Guidelines on the definition of an artificial intelligence system The European Commission’s non-binding guidelines quoting Article 3(1) of the AI Act, listing its seven elements and giving examples such as spam detection and fraud detection, with the 2 February 2025 application date of the definition.
Ethical Issues in Advanced Artificial Intelligence Nick Bostrom’s 2003 paper, the source for his definition of superintelligence and for his examples of what does and does not qualify, including Deep Blue and corporations.
Ban Artificial Superintelligence Act release summary The sponsors’ summary of the bill, giving its two-part definition of artificial superintelligence, the proposed prohibition and pause, and the penalties of up to 20 years in prison and a corporate death penalty.
What’s in a Name? Executive Order Directs Agencies to Say “Super Intelligence,” Not “Artificial Intelligence” A law firm analysis dated 13 October 2026 covering the order’s lack of statutory change or penalty, the Federal Register citation, the NIST notice, the 8 October Truth Social post and the White House Accord signatories.
AI by any other name? White House directs federal agencies to enter the era of ‘super intelligence’ An IAPP analysis of 30 September 2026 describing the order as a linguistic change that leaves existing legal definitions intact and creates two parallel vocabularies for governance programmes.
Trump renamed AI to ‘super intelligence.’ That label means something already. A PolitiFact explainer of 7 October 2026 supplying the president’s 29 September remark, Bostrom’s origin of the term, company usage, post-order industry behaviour, expert comments and the congressional bills and poll cited.
Dartmouth workshop A reference summary of the 1955 proposal and 1956 workshop, including the four proposers, the rival names of the period and McCarthy’s reasons for choosing the word artificial.
Synthetic intelligence A reference summary of Haugeland’s diamond analogy and of the philosophical dispute between McDermott, Searle, Dennett, Russell and Norvig over whether machine intelligence is real or simulated.
Synthetic Intelligence (P2P Foundation Wiki) A compilation of Benjamin Bratton’s writing on synthetic intelligence, including his use of Herbert Simon’s artificial and synthetic distinction and his second sense of human and machine synthesis.
| Citing this article? Brief excerpts are welcome. Please credit Webiano.digital, name the author where stated, and include a link to https://webiano.digital and to this original article. Full or substantial republication requires prior written permission. Read our Copyright and Content Use Policy. |
This article was prepared with the assistance of artificial intelligence tools. The content underwent expert human review, and Webiano Digital & Marketing Agency assumes editorial responsibility for its final version and publication.
