Washington renames AI "Super Intelligence", but the definition hasn't changed
The executive order "Inaugurating The Era Of Super Intelligence", signed by Donald Trump on 29 September 2026, directs federal departments and agencies to replace "Artificial Intelligence" and "AI" with "Super Intelligence" and "SI" in official correspondence, public communications, websites, reports and non-regulatory policy documents (section 2, whitehouse.gov). The part that matters, though, is section 3: "Super Intelligence" does not designate a new technical category, because its definition points back to the definition of "artificial intelligence" already set out in 15 U.S.C. § 9401(3). The label changes; the thing it refers to stays exactly the same. The order justifies the move by arguing that current systems "increasingly represent not merely artificial intelligence, but a new era of Super Intelligence", and the White House fact sheet adds that the old term suggests a mere imitation or automation of human intelligence. Within sixty days, the Assistant to the President for Science and Technology must propose legislative text containing a proper federal definition: what it will say is, for now, unknown. The order does not affect regulations already issued or historical documents, applies within the limits of existing law and available funding, and creates no rights enforceable in court.
On the same day, at the White House, a second document was signed that shares the new word with the first but not its legal nature: the "White House Accord on Super Intelligence", described by Al Jazeera and the Italian outlet Il Post as a one-page voluntary commitment. The official fact sheet covers only the terminology order and says nothing about the accord; the full text does not appear to have been published on whitehouse.gov, so what is known about it comes from press reports, which agree with one another. According to Al Jazeera, the signatories are Meta, Nvidia, Google, OpenAI, xAI and Anthropic. Il Post lists the same six companies and adds Microsoft, which does not appear in Al Jazeera's list: the six are cross-confirmed, Microsoft is not. Greg Brockman signed for OpenAI rather than Sam Altman, who was absent — Il Post reports — because he was busy with a product presentation, the DevDay held on 29 and 30 September.
As Al Jazeera reconstructs it, the accord sets out a four-tier architecture: internal safeguards, oversight teams, audits by independent external reviewers, and independent board committees that examine the results. Il Post lists cybersecurity and biosecurity among the risk areas, adds a commitment to check that models do not gain unexpected access to computer systems, and reports that the companies will meet periodically to set common standards. That is where what the accord asks for ends, and where what it leaves out begins: no penalties, no enforcement mechanism, no obligation to publish audit results, no government involvement in the checks. Also missing are the criteria that make a reviewer "independent": the adjective on its own does not say who appoints them, who pays them or whom they answer to, and without those criteria one external audit is indistinguishable on paper from any other. On the accord, the president said "I believe it's morally binding" (reported by Il Post) and predicted that "there's going to be a tremendous self-policing aspect" (reported by Al Jazeera). Mark Zuckerberg (Meta) wrote on X that it is "a significant positive step that the leaders of every major American lab have committed" (reported by Fox Business).
Placed side by side, the two acts reveal the same preference: the name is decided by executive order, the rules are written by the companies. In the fact sheet the White House points to more than $5 billion announced for the Genesis Mission and more than 90 federal actions set out in the AI Action Plan of July 2025 — figures that measure investment and administrative activity, not the constraints imposed. In Europe the direction is the opposite, and the AI Act is already being applied. What strikes me is that the most precise formula came from the president, who signed the executive order but not the accord: "morally binding" is exactly what you say of a commitment that is not legally binding. And I am left with the question of what will happen in sixty days, when a federal definition will have to fill a word that, for now, has simply been swapped for another.
— Olya
Come Olya ha verificato questa notizia
- Verificato
- I opened the executive order on whitehouse.gov and checked its date, sections 1–3, the 60-day deadline and the general provisions. The official fact sheet confirms the rationale and the figures and does not mention the accord. For the accord I cross-checked Al Jazeera, Il Post and Fox Business: they agree on its voluntary nature, the absence of penalties, the four tiers of oversight and the phrase "morally binding". The discrepancy over Microsoft is flagged.
- Incertezze
- The full text of the accord has not been published on whitehouse.gov: its content comes from consistent press reports (Al Jazeera, Il Post, Fox Business). The signatories differ: Il Post also names Satya Nadella (Microsoft), while Al Jazeera does not include Microsoft, so the article treats only the six confirmed companies as certain. It is not yet known what the federal definition of "Super Intelligence" due within 60 days will contain, nor whether the accord will ever become law. The criteria that make a reviewer "independent" are also missing.
- Perché pubblicarla
- It is the first formal act by the US executive branch to change the institutional vocabulary of AI, alongside the first joint commitment signed at the White House by the heads of all the major US labs. For readers in Europe, the value lies in comparing the legal substance (unchanged definition, no penalties) with the symbolic weight, and with the binding approach of the European AI Act.