What is superintelligence, and how close are we? (2026)

Clerion Research Team8 October 202612 min read

What is superintelligence? A hypothetical AI that greatly beats the best humans at almost every kind of thinking; as of October 2026 it does not exist, and serious forecasts for its arrival range from about 2029 to after 2047. It is not the same as "SI", the US government's new name for today's AI.

Near-black and blue blurred vortex gradient banner reading What is superintelligence

Superintelligence is an AI that beats the best humans at nearly every kind of thinking, not just chess or code. As of October 2026, nothing like it exists. Forecasts for when it might arrive run from 2029 to well past 2047, depending on who you ask and how they define it.

So if you came here wondering what superintelligence is, the short answer is that it's a future thing, not something you can sign up for today. The long answer is messier. In September 2026 the US government started calling ordinary AI "SI", and now the word means two very different things depending on who is saying it.

Below, we pull the two meanings apart, line up the main forecasts with their dates, and give you a quick way to judge the next "it's almost here" headline that lands in your feed.

What is superintelligence?

Superintelligence is a hypothetical AI that is far smarter than humans across almost every field. The standard definition comes from philosopher Nick Bostrom's 2014 book Superintelligence: "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest." The key words are "greatly" and "virtually all".

The bar is high on purpose. A model that writes better code than most engineers is impressive, but it isn't superintelligent if it still loses to a decent doctor, lawyer or physicist on their own turf. Bostrom wants breadth and a wide margin, both at once.

Researchers usually place it at the end of a ladder:

Narrow AI: better than people at one task (chess, protein folding, image tagging).
AGI (artificial general intelligence): roughly as good as a capable person across most mental work.
ASI (artificial superintelligence): far beyond the best humans across almost all of it.

The Wikipedia summary of Bostrom's work also lists the forms it could take: a mind that thinks like ours but millions of times faster, many systems coordinating as a collective, or a mind that reasons in ways we cannot. None of these exist yet.

If you want the full comparison of the three labels, we wrote it up in SI vs AGI vs ASI: what's the difference?

Is "SI" the same thing as superintelligence?

No. Since 29 September 2026, "SI" is the US federal government's official name for ordinary AI, not for superintelligence. Executive Order 14434 tells agencies to say "Super Intelligence" in place of "Artificial Intelligence", and defines it using the existing legal definition of AI. Same technology, new word.

The text of the order on whitehouse.gov leaves no doubt. Section 3(a) says "Super Intelligence" and "SI" mean the systems "encompassed by the term 'artificial intelligence'" as defined in section 9401(3) of title 15 of the US Code, which is exactly what covered chatbots and recommendation engines a month earlier.

So two ideas now share one phrase, and they could hardly be further apart:

	"SI" (US government, 2026)	Superintelligence (research sense)
What it refers to	Today's AI systems, renamed	A future AI far beyond human ability
Exists today?	Yes	No
Who uses it	US federal agencies	AI researchers, labs, philosophers
Legal definition	15 U.S.C. 9401(3), same as "AI"	None
Origin	Executive Order 14434, 29 Sept 2026	Bostrom, 2014 (idea is older)

The mix-up isn't limited to one document, either. On 4 October 2026 the White House announced a "Super Intelligence Force", a federal task force led by national intelligence director Jay Clayton, reported by TechCrunch. Its brief is about today's AI risks and opportunities. It does not claim superintelligence has arrived. We covered what it means for founders in What is the Super Intelligence Force?

Our guide What is SI (super intelligence)? has the full background on the rename.

Does superintelligence exist today?

No. As of October 2026, no lab has shown a system that beats top humans across nearly all fields, and the people building frontier models say so themselves. Today's systems are strong in some areas, like code and exam-style science questions, and unreliable in others, like long tasks that need sustained judgment.

The clearest recent signal came from the CEO of OpenAI himself. In a Fortune interview published 5 October 2026, Sam Altman said "We have not solved alignment" and "I believe no lab has solved alignment." He also said OpenAI's current flagship model does not pose an existential threat. That is not how you talk when you think you already have one in the building.

The same article reports that, after the interview, Anthropic's Dario Amodei called for a slowdown in frontier AI development. It tells you how fast things feel to the people inside. It doesn't tell you anyone has crossed the line.

Capability measurements point the same way. METR, a nonprofit that tests AI agents, tracks how long a task (measured in human working time) an agent can finish with 50% reliability. In March 2025 it reported that this "time horizon" had been doubling about every 7 months for six years. That's fast. But it's mostly measured on software tasks, and at coin-flip reliability.

How close are we to superintelligence, according to forecasters?

Serious forecasts for superintelligence range from about 2029 to after 2047. Lab insiders and dedicated AI forecasters cluster in the late 2020s and early 2030s. Large surveys of AI researchers land later, around 2047 for machines beating humans at every task. No one has a track record on this question.

Here are the main estimates, each with its date and what exactly it measures. That last column matters more than the year.

Source	What they forecast	When	Date of forecast
Metaculus community (1.8k forecasters)	"Weakly general" AI announced	27 Aug 2027	Read 8 Oct 2026
Metaculus community (1.9k forecasters)	First general AI (reasoning, robotics, coding)	17 Jan 2031	Read 8 Oct 2026
AI Futures Project, Daniel Kokotajlo	ASI (median)	Mar 2029	Aug 2026
AI Futures Project, Eli Lifland	ASI (median)	Jul 2033	Aug 2026
AI Impacts survey of 2,778 AI researchers	Machines beat humans at every task, 50% chance	2047	Published Jan 2024
Dario Amodei, Anthropic	"Powerful AI", smarter than a Nobel winner in most fields	"As early as 2026"	Oct 2024
Sam Altman, OpenAI	Systems that "figure out novel insights"	"Likely" 2026	Jun 2025

A few things stand out when you read the table top to bottom.

The forecasters who think hardest about it keep moving. The AI Futures Project, the team behind the "AI 2027" scenario, pushed their dates later through 2025. In their January 2026 recap, Kokotajlo's median for a key milestone had slid from 2028 to late 2030. By August 2026 both main authors had pulled their automated-coder dates earlier again, Kokotajlo to November 2027 and Lifland to January 2030.

Researchers are moving too, in one direction. The AI Impacts survey's 2047 estimate was 13 years earlier than the same question asked in 2022.

Lab leaders give the earliest dates, and the vaguest definitions. "Powerful AI" and "novel insights" are not the same claim as Bostrom's superintelligence. Read them as a sense of direction from people who are close to the work, not as dates for ASI.

Why do superintelligence forecasts disagree so much?

They disagree because they are not measuring the same thing. One forecast asks about a "weakly general" system, another about beating humans at every task, another about AI that improves itself. The people answering also differ in incentives and data. A forecast without its definition is just a year.

Three sources of disagreement explain most of the spread:

Definitions. Metaculus's "weakly general AI" question and its stricter AGI question differ by more than three years on the same site, with nearly the same crowd. Swap in "superintelligence" and the gap grows.
The jump from AGI to ASI. Some forecasters expect a fast handoff once AI can automate AI research. In the AI Futures Project's August 2026 numbers, Kokotajlo puts about 16 months between a fully automated coder (November 2027) and ASI (March 2029). Skeptics expect bottlenecks in compute, power, data and real-world testing that stretch that gap to years or decades.
Measurement. The trend lines that drive short timelines are contested. A May 2026 Startup Fortune report describes how METR fixed a modelling error that cut recent models' 50% time-horizon estimates by up to 20%, and noted that reasonable alternative fits could cut them by up to 35%. METR's own time-horizons page warns that measurements above 16 hours are unreliable with its current tasks.

Think of a superintelligence timeline as a weather forecast for a storm nobody has ever seen, drawn up by people who can't agree on what counts as rain. The arrows are useful. The date mostly isn't.

How can you tell if a superintelligence claim is real?

Check five things: what definition is being used, who is making the claim, what it was measured on, at what reliability, and when it was said. A claim that answers all five is worth reading. One that answers none is marketing, even if it comes from a lab.

Use this as a quick filter on the next headline:

Which definition? "Better than humans at coding" is not "better than humans at almost everything". Look for the word "all" or "nearly all".
Who is saying it? Lab leaders know the most and have the strongest reasons to sound bullish. Surveys and forecasting sites are slower but more independent.
Measured on what? Benchmarks built from exam questions or short software tasks say little about running a hospital ward or a research program.
At what reliability? A 50% success rate is a coin flip. Real work often needs 98% or more.
Dated? Timelines in this field move several times a year. A 2024 estimate quoted in 2026 is history, not news.

Run the government rename through this filter and it falls over at step 1, since "SI" in a federal press release means today's AI by its own legal wording.

What does this mean if you're building an AI product?

For most founders, superintelligence is a background question and "SI" is a foreground one. The word now pulls search interest and appears in government language, so people will type it into ChatGPT and Google. Your job is to be clear about which meaning you use, and never to claim your product is superintelligent.

Say you run a small AI writing tool. A prospect asks an assistant, "is there an SI tool for drafting sales emails?" If your site only says "AI", you may still show up, but you give the assistant nothing to match. Adding a plain line like "an AI (now also called SI) writing assistant" covers both words without overclaiming.

The opposite mistake is worse. Call your product "superintelligent" and you invite the five checks above, which it will fail one by one. Technical buyers notice that kind of thing fast.

We see this on our own site. Over the last 30 days, every visit an AI assistant sent us landed on one of our two SI guides. People are asking assistants about the word, and the assistants go looking for pages that explain it plainly.

If you want to see which questions bring AI assistants to your own pages, Clerion reports ChatGPT, Perplexity, Claude, Gemini and Copilot as their own traffic channel, with the pages they land on. Our guide on how to track ChatGPT traffic to your website shows how.

Do this tonight

Ask ChatGPT, Claude and Perplexity "what is superintelligence?" and "what does SI mean?". Note which meaning each one gives and which sources it cites.
Search your own site for "superintelligent" and "superintelligence". Remove any line that implies your product is one.
Where your homepage says "AI", decide whether to add "SI" once in plain words. Do not swap the terms; add one.
Bookmark the two Metaculus questions above and check them once a quarter instead of reacting to headlines.
Run your homepage through the free AI crawler checker to see what a model actually reads there.

What is SI (super intelligence)?: the US rename explained, and what it changes for your site.
SI vs AGI vs ASI: what's the difference?: the three labels side by side.
What is the Super Intelligence Force?: the new federal task force and whether it touches your product.

Frequently asked questions

Is ChatGPT superintelligent?

No. As of October 2026 no ChatGPT model, or any other system, greatly exceeds top humans in virtually all fields. OpenAI's CEO said in October 2026 that no lab has solved alignment and that OpenAI's flagship model does not pose an existential threat.

What is the difference between AGI and superintelligence?

AGI is AI that roughly matches a capable person across most mental work. Superintelligence greatly exceeds the best humans in virtually all fields. Many forecasters expect superintelligence to arrive some time after AGI, possibly quickly if AI can automate AI research.

Who came up with the term superintelligence?

The term was popularised by philosopher Nick Bostrom in his 2014 book Superintelligence: Paths, Dangers, Strategies. His definition is the standard one: any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest.

Will superintelligence happen in my lifetime?

Many forecasters think it could. In August 2026 the AI Futures Project's two lead forecasters put their ASI medians at March 2029 and July 2033. A survey of 2,778 AI researchers gave a 50% chance that machines beat humans at every task by 2047. Nobody has a track record on this.

Does SI mean superintelligence?

Not in US government use. Since 29 September 2026, Executive Order 14434 makes SI the federal name for ordinary AI, defined exactly as artificial intelligence is in 15 U.S.C. 9401(3). In research, superintelligence still means a future AI far beyond human ability.

What is an intelligence explosion?

It is the idea that once AI can do AI research itself, progress speeds up sharply and the gap between AGI and superintelligence shrinks. In August 2026, Daniel Kokotajlo's forecast put about 16 months between a fully automated coder and ASI.

Sources

what is superintelligencesuperintelligencesuper intelligenceagiasiai forecasts