Superintelligence: news, explainers and who is building it

What superintelligence means, who says they are building it, and what changed this week — sourced news and explainers.

Updated 22 October 2026

Superintelligence is a hypothetical intelligence, usually imagined as a machine, that is far more capable than the best human minds at almost every kind of thinking. A widely quoted definition comes from the philosopher Nick Bostrom's 2014 book Superintelligence: Paths, Dangers, Strategies (chapter 2), as quoted by Wikipedia: "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest". The word is easy to define and hard to pin down, because the people who say they are building it rarely use it on its own. Elon Musk and Sam Altman say digital superintelligence, Safe Superintelligence Inc. says safe, Meta says personal and Microsoft AI says humanist. Meta sets its personal approach against superintelligence "directed centrally", and Microsoft AI wants systems that work "in service of" people; the table further down sets out who uses which phrase, in their own words.

What superintelligence means

The idea is older than the word. In a paper published in 1966, the statistician I. J. Good described an "ultraintelligent machine" that could design even better machines, setting off what he called an "intelligence explosion". Bostrom gave the term an influential definition in his 1998 paper "How long before superintelligence?", where a superintelligence is "an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills". His 2014 book tightened that into the definition quoted above.

Three points follow from those definitions. First, the bar is breadth, not a single skill: a system that beats every human at chess, or at writing code, is not superintelligent unless it leads across almost every field. Second, the definition says nothing about how it is built; in principle it could be software or an enhanced human mind. Third, it is a description, not a test. There is no agreed benchmark that would show it has arrived, which is why claims about it deserve close reading. The explainer covers what superintelligence means in full.

Superintelligence vs AGI vs ASI

The three terms are often used as if they were the same, but they set different thresholds. Artificial general intelligence (AGI) is, in Wikipedia's summary, AI that "matches or surpasses human capabilities across virtually all cognitive tasks". Matching people is the bar. The same article names OpenAI, Google, SpaceXAI and Meta as companies with AGI as a stated goal.

Superintelligence sets the bar much higher: greatly exceeding humans, not keeping pace with them. Artificial superintelligence (ASI) is simply superintelligence in a machine, so in practice ASI and superintelligence usually mean the same thing. Safe Superintelligence Inc. and Microsoft's MAI Superintelligence Team put "superintelligence" in their names, while SoftBank Group calls its goal "Artificial Super Intelligence (ASI)". The usual story connects them: a system reaches AGI, then improves itself or is scaled up until it passes far beyond human level. The side-by-side comparison of AGI and ASI has the labs' own wording for each.

Which phrase does each group use?

Each group adds its own adjective, and each choice says something. Musk has used "digital superintelligence" since at least August 2014, and in July 2023 defined it as AI smarter than any human at anything (CNBC, Fox Business); Sam Altman used the same phrase in "The Gentle Singularity". Safe Superintelligence Inc., founded by Ilya Sutskever, Daniel Gross and Daniel Levy, puts safety in its name and says it has "one goal and one product". Meta's "personal" framing is about a superintelligence for each individual rather than one system run centrally, in Mark Zuckerberg's letter of 30 July 2025. Microsoft AI, led by Mustafa Suleyman, says "humanist" to stress a system that stays in service of people.

Why 2026 is different

Until this year, superintelligence was mostly a word used by philosophers, researchers and AI labs. On 29 September 2026 "Super Intelligence" became US government vocabulary. Executive Order 14434, "Inaugurating the Era of Super Intelligence", says that the executive branch "shall use the terms ‘Super Intelligence’ and ‘SI’ in place of ‘Artificial Intelligence’ and ‘AI’" in official correspondence, websites, reports and policy documents, to the maximum extent permitted by law. The order does not claim that superintelligence exists. For now, SI simply means what US law already defines as artificial intelligence, and the order asks for proposed legislative language on a new federal definition within 60 days. The result is two very different meanings in circulation: Bostrom's, which describes something no one has shown, and Washington's, which covers AI systems in use today. Reading any 2026 headline about "SI" means checking which one is meant.