What is superintelligence? The definition, who claims it is near, and why Washington just adopted the word

What is superintelligence? In the strict sense it is a hypothetical mind that is far better than any human at nearly everything that counts as thinking. Nothing like it exists today. Yet the word is suddenly everywhere: Meta says it is "in sight," a startup founded around it says it is "within reach," and on September 29, 2026 the White House ordered federal agencies to start calling AI "Super Intelligence." This guide separates the definition from the sales pitch.

The definition that most people mean

The philosopher Nick Bostrom, defines superintelligence as "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest." That is the wording Wikipedia quotes, and it is the standard reference point. Wikipedia's own lead is looser: "a hypothetical agent that possesses intelligence surpassing that of the most gifted human minds."

Two details matter. First, the bar is not "better than a person at chess," which machines cleared decades ago. It is better across virtually all domains at once. Second, the definition says nothing about how the mind is built. It could be one model, a swarm of cooperating models, or something else entirely.

Bostrom described three routes to get there, as Wikipedia summarizes them. Speed superintelligence thinks like a human but far faster, for example an emulated mind on faster hardware. Collective superintelligence is many reasoning systems coordinating well enough to beat any individual. Quality superintelligence simply reasons better than any human can.

Where it sits on the ladder

Superintelligence is the top rung of a ladder, not the first. Narrow AI is built for a task. Artificial general intelligence, or AGI, is described on Wikipedia as a system that "matches or surpasses human capabilities across virtually all cognitive tasks." Artificial superintelligence, or ASI, is the step beyond. Google DeepMind researchers made the ladder explicit in a paper on levels of AGI, which reserves its top tier, "Level 5: Superhuman," for artificial superintelligence. Our companion explainer, AGI vs ASI, walks through the glossary.

Who is using the word, and for what

The word now does different jobs depending on the speaker.

• OpenAI. Sam Altman wrote in "The Gentle Singularity" in June 2025: "We are past the event horizon; the takeoff has started. Humanity is close to building digital superintelligence, and at least so far it's much less weird than it seems like it should be." It is a company claim about direction, with no date attached to the arrival itself.
• Meta. Mark Zuckerberg's July 30, 2025 letter says "Developing superintelligence is now in sight," and pitches "personal superintelligence" that helps individuals, rather than central automation.
• Safe Superintelligence Inc. The startup's site says "Building safe superintelligence (SSI) is the most important technical problem of our time" and "Superintelligence is within reach."
• Anthropic. Dario Amodei avoids the word in his January 2026 essay "The Adolescence of Technology" and uses "powerful AI" instead: a model "smarter than a Nobel Prize winner across most relevant fields." He writes it could be as little as one to two years away. That is a forecast, and it describes something narrower than Bostrom's bar.

Notice what these have in common: each is a statement from a company or its chief executive about its own field. None is an independent measurement.

What the evidence actually says

No public benchmark certifies superintelligence. The nearest thing to independent evidence is how researchers themselves forecast. The 2023 AI Impacts survey of 2,778 authors who had published at top AI venues gave a 50% chance that unaided machines outperform humans at every task by 2047, which was 13 years earlier than a similar survey a year before. It also found between 38% and 51% of respondents gave at least a 10% chance to outcomes "as bad as human extinction," depending on how the question was framed. That survey is nearly three years old, and the labs say the pace has changed since, so read it as a baseline and not a verdict. Note too that it measures human-level performance on every task, which is a lower bar than superintelligence.

The AI Futures Project, whose April 2025 scenario "AI 2027" put artificial superintelligence in December 2027, has itself pushed its median dates back; its authors noted in November 2025 that 2027 was their most likely single year, but their medians were somewhat longer. We cover the forecasts in the AGI timeline tracker.

The rename: a word, not a capability

On September 29, 2026 President Trump signed an order titled "Inaugurating the Era of Super Intelligence." It tells the executive branch to use "Super Intelligence" and "SI" in place of "Artificial Intelligence" and "AI" in official correspondence, public communications, websites, reports and other non-statutory documents, to the maximum extent permitted by law. The legal definition did not move: Section 3(a) keeps the existing statutory definition of artificial intelligence at 15 U.S.C. 9401(3) until a later presidential action or an Act of Congress changes it. The science adviser has 60 days to propose legislative language. Our full explainer is here, and the companion industry pledge is covered in the accord tracker.

So the US government now uses the word without having defined it in law, which is a curious place to be.

Our take

Use the word precisely. It describes a hypothetical system defined by a very high bar, and today it works mainly as a destination companies name in their mission statements and, since September 29, as a federal style rule. When someone says superintelligence is near, ask four things: what test, who ran it, what date, and what happens if it is wrong. Until a claim answers those, file it under forecast or branding. We will update this page as the definition is written into law or as an independent evaluation says otherwise.