Superintelligence: who is using the word, and what they mean

A dated snapshot of the U.S. government rename, how AI companies use the term, the competing definitions, what forecasts, polls and critics say, and what is still open, with links to every source.

Updated November 2, 2026. We plan to refresh this page monthly. This is research-based, not our own testing or fieldwork: we read the sources linked below. Forecasts, company claims and opinions are labeled as such, and we take no side on whether superintelligence is near or should be banned. Items marked "Reported" come from one or a few outlets, are statements by the people quoted, or are allegations, and we have not confirmed them.

Where things stand

The word is now in U.S. government documents

Executive Order 14434, signed September 29, 2026 and published October 2, tells federal agencies to use "Super Intelligence" and "SI" instead of "Artificial Intelligence" and "AI" in documents that are not statutes. For now the term means whatever the existing legal definition of AI covers, and the order gives the White House science office 60 days to propose a federal definition. Law-firm analysis notes the order has no penalties and covers only the executive branch. On October 4 the president announced a "Super Intelligence Force", which TechCrunch describes as a rebranding of AI. [White House, Arnold & Porter, TechCrunch]

Companies use the word for different things

Meta's Superintelligence Labs, led by Alexandr Wang, builds the Muse models; Meta's founder framed the goal as "personal superintelligence", and Meta guides to $130 billion to $145 billion of capital spending for 2026. Microsoft AI formed a "MAI Superintelligence Team" in November 2025 around what its chief calls humanist superintelligence: domain-specific systems with limits on autonomy. At an India summit in February 2026, OpenAI's Sam Altman said OpenAI may be "a couple of years away from early versions of true superintelligence". Anthropic's CEO describes "powerful AI" and "a country of geniuses in a datacenter" instead. Safe Superintelligence Inc. says its goal is a "safe superintelligence". [Fortune on Meta Superintelligence Labs, Meta, Microsoft AI, MediaNama on Altman at the India AI Impact Summit, Dario Amodei, TechCrunch]

There is no agreed definition

Philosopher Nick Bostrom defined superintelligence as an intellect that greatly exceeds human performance in virtually all areas of interest. OpenAI's charter describes AGI as highly autonomous systems that outperform humans at most economically valuable work. A 2024 Google DeepMind paper rates AI on performance and generality, and its top level, "Superhuman", means outperforming all humans. Meta's lab name and the federal rename come with no technical definition. So "superintelligence" can mean a research goal, a company brand or a government label, depending on who says it. [Wikipedia, OpenAI Charter, Google DeepMind]

Claims run ahead of what is measured

METR measures how long a task an AI can complete at a given reliability. Its January 2026 note put Claude Opus 4.5's 50% time horizon at about 4 hours 49 minutes, with an error range of roughly 1 hour 49 minutes to 20 hours 25 minutes, and warns that a 50% horizon is not safe delegation and differs by domain. In May 2026 The Decoder reported that an early Claude Mythos Preview scored at least 16 hours, which METR called near the upper end of what its task set can measure. A time horizon is a narrow measure of task length, not of general intelligence. [METR, The Decoder on METR and Claude Mythos]

Forecasts, polls and critics

What lab leaders forecast

These are forecasts, not results. Google DeepMind's Demis Hassabis said AGI could be two to three years away, and acknowledged some of his urgent language was meant to get governments and the public to take the technology seriously. Altman's February 2026 remarks put early superintelligence a couple of years out. Amodei wrote in January 2026 that powerful AI could be as little as one to two years away, and in a later interview gave about 90% confidence in a "country of geniuses" within ten years. [Tom's Guide via Yahoo, MediaNama on Altman at the India AI Impact Summit, Dario Amodei, Dwarkesh Patel]

What expert surveys show

The AI Impacts survey of 2,778 researchers in October 2023 gave a 50% chance of "high-level machine intelligence", meaning unaided machines doing every task better and more cheaply than people, by 2047; it is three years old. The Forecasting Research Institute's wave 8, fielded April 20 to May 11, 2026 with 205 experts, 52 superforecasters and 601 members of the public, found median AGI years of 2050 for experts, 2047 for superforecasters and 2055 for the public, under a narrow, conditional definition. Both give later dates than the lab leaders above. [AI Impacts, Forecasting Research Institute]

What polls show

A Politico/Public First poll of 2,064 U.S. adults on September 13 to 15, 2026 (margin of error 2.2 points) found 17% saw destruction of humanity as almost certain, 20% as a significant risk and 26% as a moderate risk; 48% backed a pause and 31% favored continued development, and 26% called AI leaders' warnings marketing. We have not seen the question wording. A poll the Future of Life Institute sponsored in 2025 found 64% of 2,000 U.S. adults said superhuman AI should not be developed until proven safe, or never; its sponsor campaigns for a ban. [Politico/Public First poll, Future of Life Institute]

What skeptics argue

These are opinions, quoted as theirs. Stanford's James Landay says boosters keep shifting the AGI definition and inflate evaluations to attract investment. Cornell's John Thickstun says the terms are ill-defined and confident predictions are hubristic. Economist Servaas Storm argues that scaling language models cannot produce intelligence and that the claim props up an investment boom. Researchers Arvind Narayanan and Sayash Kapoor argue AI behaves like a normal general-purpose technology whose effects arrive gradually. [Yahoo Finance, Institute for New Economic Thinking, The Ringer]

Reported lately

Newest first. Dates are when the news was reported.

Oct. 22, 2026Reported

Report of Altman's "three endgames" (allegation)

Forbes relayed a report by The Information, based on anonymous sources, that Sam Altman described three possible end states for the race to superintelligence: an international body like the IAEA, a CERN-style merger of labs, or a first mover starving rivals of chips. Forbes did not confirm it, and we have not seen OpenAI comment. [Forbes on a report about Altman's three endgames]

Oct. 5, 2026Reported

Altman interview on control

In an interview reported by Fortune, Altman said humans can stay in control of AI, including superintelligent AI, by staying a few steps ahead of its capabilities, and that no lab has fully solved alignment. He said a 10% chance of AI killing everyone by the end of the decade is unacceptable. [Fortune]

Oct. 4, 2026Reported

Super Intelligence Force, and a company rename

The president announced a "Super Intelligence Force" in a post; TechCrunch presents it as a rebranding of AI. Fox Business reported that Elon Musk confirmed on X that SpaceXAI will be renamed SpaceXSI, writing "No more AI. SI, it's better." We did not read the posts themselves. [TechCrunch, Fox Business]

Sept. 29 to Oct. 2, 2026Reported

Executive Order 14434

The order was signed September 29 and published October 2 in the Federal Register. It gives the Office of Science and Technology Policy 60 days to propose a federal definition of "Super Intelligence", which puts the deadline around the end of November. [White House, Arnold & Porter]

Sept. 23, 2026Reported

Sanders and Casar bill to ban superintelligence

Senator Bernie Sanders and Representative Greg Casar unveiled a bill that would permanently ban superintelligence, pause the most advanced AI until a new Department of AI sets safety rules, and require federal approval before deployment. The AP says it faces long odds in the Republican-controlled Congress. [PBS NewsHour]

Sept. 8, 2026Reported

UK Artificial Superintelligence Bill

Labour MP Alex Sobel introduced a private member's bill that would prohibit developing, deploying and operating artificial superintelligence; its second reading is scheduled for November 13, 2026. IBTimes UK reports the government rejected the bill's approach while exploring narrower measures, which we found in that outlet only. [LexisNexis UK, IBTimes UK]

July 27, 2026Reported

Safe Superintelligence and Nvidia partner

Ilya Sutskever's Safe Superintelligence Inc. announced a long-term partnership with Nvidia for access to its Vera Rubin systems. The deal amount was not disclosed; TechCrunch cites a source and a Bloomberg report with figures we could not confirm. [TechCrunch]

July 17, 2026Reported

Xi Jinping refers to loss of control

The MERICS think tank says Xi referred to "loss of control" of AI in a speech at the World AI Conference in Shanghai, which it calls his first known public reference to the risk. MERICS adds it is unclear whether Beijing shares Western alignment concerns. [MERICS]

Nov. 6, 2025Reported

Microsoft AI forms a superintelligence team

Microsoft AI chief Mustafa Suleyman announced the MAI Superintelligence Team and defined "humanist superintelligence" as advanced AI built around specific domains, with limits on autonomy. He said he expects medical superintelligence in the next few years, which is a forecast, and cited an 85% score on a medical case benchmark, which is a company claim. [Microsoft AI]

Oct. 22, 2025Reported

Statement on Superintelligence

The Future of Life Institute published a statement calling for a prohibition on developing superintelligence until there is broad scientific consensus it can be done safely and strong public support. CyberScoop reported more than 700 signers at launch, including Yoshua Bengio, Geoffrey Hinton and Steve Wozniak. The campaign's site now shows 78,115 signatures, a self-reported figure with no stated verification or date. [CyberScoop, superintelligence-statement.org]

Open questions

What will "Super Intelligence" mean in federal documents?

Until the science office proposes a definition, around the end of November, the order says the term covers what the legal definition of AI covers. We do not know whether the proposal will describe a technical level or keep the rename. [White House, Arnold & Porter]

Are the timelines forecasts or promotion?

Lab leaders give short timelines, expert surveys give later ones, and Hassabis acknowledged some urgent wording was meant to draw attention. Critics call the claims hype; the leaders describe them as real expectations. The sources do not settle it. [Tom's Guide via Yahoo, Forecasting Research Institute, Yahoo Finance]

How would anyone know it had arrived?

We found no accepted test. Definitions differ, and METR says its own measures are limited and differ by domain. [Google DeepMind, METR]

Will bans or rules follow?

The U.S. bill faces long odds according to the AP, and the UK bill's second reading is set for November 13. Polling shows support for a pause in one poll we read, but wording and sponsors differ. We have not found a binding rule that bans developing superintelligence anywhere. [PBS NewsHour, IBTimes UK, Politico/Public First poll]

Does the new name change anything for products?

The order has no penalties and covers federal documents. We found nothing showing that tools or prices change because of the word, so we keep describing tools by what they do. [White House, Arnold & Porter]

If you are reading or buying "superintelligence" news

Our own checklist, drawn from the reports above. It is not legal or investment advice.

When you see "superintelligence", ask who is saying it and which meaning they use: a company brand, a forecast, a legal label or a technical level.
Separate forecasts, which say what someone expects, from measurements, which say what was tested and how.
Check the date and who benefits from the claim; many items here are company statements or single-outlet reports.
A product that uses the word is still judged by its features, price and data handling, like any other tool.
If you work with U.S. federal agencies, watch for the science office's definition around the end of November; this is not legal advice.
Read the best arguments from people who disagree with you; the sources above include both ban supporters and skeptics.

Sources

We read the pages above directly or through the outlet named. If a source is wrong or out of date, tell us and we will fix it.