Executive Summary

Key Numbers

Three numbers: the deadline the order fixed for legislative language defining Super Intelligence, how many awards a keyword search for "AI funding" turns up in a year, and how many kinds of document the order says need not change.

Sources: Executive Order 14434, Federation of American Scientists, "Trust Issues".

The webinar itself was ordinary. The Directorate for Mathematical and Physical Sciences at the National Science Foundation held an online session to walk applicants through a funding opportunity in the mathematical sciences. The agency's own event listing puts it on Wednesday, October 7, at two in the afternoon, an informational webinar for a solicitation that bundles more than a dozen programs. FYI, the newsletter that reported the remark, gave no date and said only "Wednesday."

The remark FYI carried came from Junping Wang, a deputy division director for mathematical sciences. On the "AI" written into grant abstracts and related documents, he said, "your program director may work with you by changing AI to SI." In the same session he also said that AI tools inside the agency may be used to summarize parts of a submission "but not for making recommendations."

The sentence turns on its modal verb. This is an offer, not an instruction. Some abstracts will stay as they are because nobody asks, and program directors may judge the request differently from one another. The agency did not announce that it will rewrite its whole portfolio. It also means that rewritten documents and untouched ones will accumulate side by side in the same database.

Renaming is happening on more than one track. The same newsletter reported that the agency plans to rename eight proposal review panels to match new office names. That change comes from a reorganization rather than from the executive order. Behind both sits a year of criticism: science agencies issued far fewer awards in fiscal year 2026 than usual, and the agency closed the year with more than a billion dollars unspent.

Executive Order 14434 was signed on September 29 and published in the Federal Register on October 2 under the title "Inaugurating the Era of Super Intelligence." Section 1 gives the reason for the new name: the systems now in use have outgrown what the words "artificial intelligence" were first meant to cover. The instruction in Section 2(a) is one sentence long. "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."

Section 2(b) draws the line in the very next sentence. "Nothing in this section requires the alteration of previously issued regulations, Presidential actions, contracts, grants, or other historical documents." A decision not to reach backward is written into the order itself.

Section 3 fixes the meaning as well. "Super Intelligence" and "SI" refer to the same technology that 15 U.S.C. 9401(3) defines as "artificial intelligence." The name changes and the legal scope does not. That arrangement is not meant to last, though. The same section gives the president's assistant for science and technology 60 days, which runs to November 28, to deliver legislative language establishing a federal definition of Super Intelligence. That language is to cover modifying or expanding the existing statutory definition along with the conforming amendments that would follow. Once the definition widens or shifts, the matter stops being only about a name.

The order's reach is narrow too. It covers the executive branch, and terms written into statute stay as they are. What anyone outside the executive branch calls the same thing is not something this order settles.

NIST took the safe side. In its recent guidance on Super Intelligence it repeated the order's own words, that this "does not require the alteration of previously issued regulations, Presidential actions, contracts, grants, or other historical documents." That is the carve-out copied over, not a new standard the institute wrote.

Where the sentence sits makes its use clearer. The institute stamped a notice at the head of every page it renamed: "Pursuant to September 29, 2026, Executive Order 14434: Inaugurating the Era of Super Intelligence. AI has been changed to SI except when referring to previously completed work and historical documents per Executive Order Section 2 (b), this does not require the alteration of previously issued regulations, Presidential actions, contracts, grants, or other historical documents." The result is two names living on one page. The title reads "Human-Centered SI," while evaluation efforts and projects that predate the order still carry "AI" in their names, each with its own note saying the work was completed earlier. A separate page, "Note about 'Super Intelligence'," sits in the institute's computer security resource center for the same purpose. Both the changed places and the unchanged ones are marked.

The National Science Foundation went a step further. It told researchers that grant documents, which sit on the list that need not change, can be rewritten alongside everything else if they want. FYI noted that renaming existing grant materials would go beyond what the September order requires, and set the NIST case next to it for comparison. Same order, same carve-out, same legal definition, and two agencies drawing the line in different places.

Calling this a violation would be a stretch. The order does not require that historical documents be altered, and it does not forbid altering them either. The legal meaning is unchanged. So the episode does not take the shape of a legal problem. The difficulty shows up elsewhere. One agency leaves its past documents alone, another rewrites them on request, and inside that second agency the outcome depends on what a program director and a researcher decide. NIST pages carry a notice at the top saying the change happened; a grant abstract carries no such line. The name printed on the document keeps no trace of the choice.

Answering "how much AI funding went out last year" turns out to rest on a plain method. In the Federation of American Scientists report on the agency's funding for trustworthy AI, the authors used the advanced award search on the agency's website, entering terms such as AI, Artificial Intelligence, Machine Learning, ML, computer vision, NLP, and Reinforcement learning, and pulled awards from a single directorate, Computer and Information Science and Engineering (CISE). From fiscal year 2018 through fiscal year 2022 the search returned roughly 1,000 grants a year, about 5,000 across the five years.

The authors said plainly that the set was over-inclusive. Many grants in it were not focused on AI, because they aimed for high recall rather than high precision. Having chosen first not to miss anything, they drew 100 awards at random from each fiscal year, 500 in all, and had two people read the titles and abstracts and classify them again. What matters here is that this method is closer to standard than to exceptional. Anyone counting funding by topic in a grants database ends up counting words in abstracts.

Those 5,000 awards are also not the whole agency. The search covered one directorate, CISE. The offer to rename came from a different one, the Directorate for Mathematical and Physical Sciences (MPS), whose awards were never in that set to begin with. How far the phrase "AI funding" reaches already varied with who was counting, well before any name changed.

Now the word itself is moving. Why not add "SI" and "Super Intelligence" to the search terms, someone might ask. Adding two terms is the easy part. Timing is the hard part. If one award's abstract read "AI" in 2025 and became "SI" on some day in 2026, which year's AI funding is that award? When a database shows only the current state and keeps no record of when and how far a change went, anyone recounting 2025 sees the past through the name it was given afterward.

This change is also neither complete nor a regular substitution. The agency's offer is optional, and other agencies are not touching their records at all. So rewritten and untouched records end up in the same table. Count only "SI" and the number comes out low; count only "AI" and the rewritten ones drop out; count both and you have to check whether the same award was counted twice. What a full sweep would settle with a single substitution rule becomes, under selective change, a record-by-record reconciliation.

The mixing does not stay inside the agency either. Because the order reaches only executive branch documents, the university research administration records, paper acknowledgments, state government filings, and corporate internal policies that point at the same award keep saying "AI." The privacy professionals' association IAPP made this point in its analysis, observing that organizations will see "SI" in federal material while statutes, state law, contracts, internal policies, and vendor checklists keep showing them "AI." For anyone trying to join the two sides by award number, the same object arrives under two names.

The remedy is not hard, either. Record the change along with the name. Leave the award identifier alone, bind the old name and the new one as aliases, and log what was changed and when. The notice NIST attached to each page is the minimal form of this: the clause it relied on, the scope of the change, and what was left alone, all in one line. The order's design, exempting historical documents, was closer to avoiding the problem altogether. An agency that decides to open the exemption owes a record in return.

News like this usually gets read as politics. From where we sit with data, something else comes into view first. A name is a description, and it is also an identifier. Search, aggregation, and comparison all hold on to that word. Changing a name is therefore a data operation rather than a document edit, and leaving each party to decide how far the change goes hands that operation to somebody else later on.

When Pebblous talks about AI-Ready Data, one condition always comes attached: the lineage has to survive even when the name for the same thing differs from source to source. This episode shows at national scale what blurs when that condition is not met. Declaring that the legal meaning is unchanged does not by itself guarantee that the past can be counted again.

A team renaming its internal vocabulary faces the same three questions. Is the scope of the change written down anywhere? Is there an alias somewhere connecting the old name to the new one? Is the date of the change recorded? All three are easy before the renaming and hard to reconstruct after it.

Thank you for reading this far. The full text of the order is at the White House, the webinar remark is at AIP FYI, and the legal commentary is at IAPP and Freshfields. If your organization has a way of recording terminology changes, we would like to hear how it works.

R.1News Coverage and Legal Commentary

• 1.FYI (American Institute of Physics). (2026). "The Week of October 12, 2026." The first report of the webinar remark. Source for Junping Wang's quoted words, the comparison with NIST's own practice, the note on renaming review panels, and the criticism of unspent fiscal year 2026 funds.
• 2.IAPP. (2026). "AI by any other name: White House directs federal agencies to enter the era of super intelligence." Legal analysis of Executive Order 14434. Source for the observation that "AI" persists outside the executive branch, in state law, contracts, and internal policy.
• 3.Freshfields. (2026). "Trump executive order mandates shift to 'Super Intelligence'." Legal commentary on the order's scope, its carve-outs, and the 60-day legislative deadline.

R.2Official Documents

• 4.The White House. (2026). "Inaugurating the Era of Super Intelligence" (Executive Order 14434). Signed September 29, 2026. Source for the text of Section 1 (purpose), Section 2(a)-(b) (scope and the historical-document carve-out), and Section 3 (fixed definition and the 60-day legislative instruction). Published in the Federal Register on October 2, 2026.

R.3Data Analysis

• 5.Alexander, L., Kaushik, D. (2023). "Trust Issues: An Analysis of NSF's Funding for Trustworthy AI." Federation of American Scientists. Source for the CISE keyword-search methodology, the roughly 5,000 over-inclusive awards across fiscal years 2018-2022, and the 500-award random reclassification figures.