Trump renamed AI "SI". What changed, and what did not?

"Super Intelligence" is now more than a phrase in an AI debate. On September 29, 2026, President Donald Trump signed an executive order directing the US executive branch to use "Super Intelligence" and "SI" in place of "Artificial Intelligence" and "AI" in specified official communications. That is a real policy change. It is not, by itself, evidence that a machine has crossed a new scientific threshold. Source

For founders and engineering teams, the distinction matters. A new name can change government language, public expectations, and vendor messaging. It cannot tell you whether a system can reliably complete your workflow, protect customer data, or operate safely with access to production tools. Those questions still need evidence.

This guide separates what Trump actually changed from what researchers mean by superintelligence, then turns the discussion into practical questions for businesses deploying AI. It does not predict a date for superintelligence or claim that the September order proves it exists.

What Trump actually said and signed

The September 29 order is titled "Inaugurating The Era Of Super Intelligence." Section 2(a) states: "To the maximum extent permitted by law, executive departments and agencies (agencies) shall use "Super Intelligence" and "SI" in place of "Artificial Intelligence" and "AI" in official correspondence, public communications, websites, reports, policy documents, and other non-statutory documents within the executive branch." That is the operative instruction, rather than a claim that every private company must rename its product. Source

The order follows Trump's September 22 remarks at the United Nations. Breaking Defense reported him saying: "In other words, welcome to the new world of super intelligence." The same report quotes his explanation that "artificial" makes the technology sound fake. This is reported speech from the UN appearance; the later signed order is the primary source for the implementation scope. Source

At the September 29 gathering, CNBC reported Trump saying, "We've changed the name officially today." The White House's accompanying fact sheet also describes the order as replacing the terminology in the executive branch. Taken together, these sources establish a naming initiative. They do not establish a new model capability or a successful superintelligence experiment. Source Source

The most important detail is the definition

Section 3(a) initially gives "Super Intelligence" and "SI" the meaning of technologies and systems already encompassed by the statutory definition of artificial intelligence in 15 USC 9401(3), except where otherwise provided by law. In other words, the new policy label initially points back to an existing AI definition. The order does not set a test requiring a system to outperform humans across every field before it receives that label. Source

The referenced statute defines AI as a machine-based system that can, for human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments. That is a broad description of AI systems, not a definition requiring universally superhuman intelligence. Read the cross-reference before treating the new terminology as a declaration that technical superintelligence has arrived. Source

There is a second step. Section 3(b) directs the Assistant to the President for Science and Technology to submit proposed legislative language within 60 days, including an assessment of whether a federal SI definition should modify, expand, or supersede the existing statutory AI definition. A requested proposal is not an enacted replacement definition. The distinction between the current cross-reference and possible future legislation should remain clear in any business briefing. Source

The order also says it does not require alterations to previously issued regulations, presidential actions, contracts, grants, or other historical documents. It is subject to applicable law and available appropriations. Businesses should not infer a blanket obligation to rewrite old agreements from the headline alone. Specific procurement or legal questions need review of the applicable documents, rather than this article as legal advice. Source

Technical superintelligence is a different claim

Researchers and AI companies have used "superintelligence" to describe a much higher capability level than ordinary AI. OpenAI's May 2023 governance essay describes it as future AI systems "dramatically more capable than even AGI." Its July 2023 Superalignment article explicitly distinguishes superintelligence from AGI to stress a much higher capability level. These are historical research statements, not evidence of a September 2026 breakthrough or a current product certification. Source Source

That produces a vocabulary collision. Under the executive order, SI initially covers the technologies in the existing legal AI definition. In technical discussion, superintelligence points to a much more demanding idea about capability. Axios identified that naming problem before the order was signed: the same word was being used for a political rebrand and a distinct research ambition. Source Source

A system being excellent at one task is not enough to settle the broader question. An impressive coding demonstration, a fast document summary, or strong performance on a benchmark can be useful evidence for a particular application. It does not automatically establish broad intelligence, safe autonomy, or reliable performance on your company's data. Keep the capability claim narrow enough that someone can actually test it.

The sources reviewed here verify the terminology change. They do not demonstrate technical superintelligence. That is a deliberately limited conclusion: it avoids claiming either that the research problem has been solved or that a future breakthrough is impossible.

Was this a new research announcement?

The September 29 order's main mechanisms are terminology changes and a proposed federal definition. Its fact sheet refers to broader administration initiatives, including the Genesis Mission, but that reference should not be mistaken for the order itself creating a new superintelligence research program. We would need to examine the underlying program documents before describing its funding, scope, or deliverables in detail. Source Source

The technical research issue is real and separate. OpenAI's 2023 governance essay called making superintelligence safe an open research question. Its Superalignment article described the difficulty of humans supervising systems much smarter than themselves. Those historical statements explain why the term carries weight; they should not be presented as an up-to-date report on that team's staffing, current commitments, or results. Source Source

What business teams should ask instead of chasing the label

Start with a measurable task

Suppose you are adding an assistant to customer support. Define success around the work: retrieving the correct policy, answering from approved sources, recognizing uncertainty, and escalating the right cases. Test with representative questions and difficult exceptions. A vendor calling the system SI does not answer those questions.

For a coding assistant, use a different test set: repository-specific changes, regression checks, dependency constraints, and review effort. Measure completed work, failures, and human corrections. Do not substitute a polished demo for evidence from the environment in which the system will run.

Separate intelligence from permission

A useful answer and an authorized action are different things. An assistant that can draft a refund explanation need not have permission to issue the refund. A system that proposes a database fix need not have unrestricted production credentials. Match tool access to the task, and make high-impact actions pass through an explicit review step.

Our engineering recommendation is to design permissions, logging, and recovery before widening autonomy. Decide which data the system can read, which tools it can call, what requires human approval, and who can stop it. This checklist is implementation advice, not a new requirement created by Trump's terminology order.

Keep a record of capability and risk

NIST's Generative AI Profile is a companion to its AI Risk Management Framework, intended to help organizations identify distinctive generative-AI risks and actions that fit their goals and priorities. It offers a more useful starting point for risk discussion than a new product label. It is guidance, not proof that a particular deployment is safe. Source

For each deployment, keep a short record of the intended use, evaluation results, known failures, permitted tools, escalation path, and rollback plan. Review it when the model, data, or workflow changes. Codesprint's website is a starting point for discussing the engineering behind those controls. The objective is useful automation with evidence behind it, regardless of the acronym on the slide.