Who is building superintelligence?

First, the distinction that matters on this page

There are three different things a company can be doing, and the September 2026 rename has made them easy to blur.

1. Building AI. Almost every technology company now does this. Under the terminology the White House published on 22 September 2026 (artificial intelligence “hereinafter officially called ‘Super Intelligence’”), all of it would be called “Super Intelligence”. That is a naming decision; it does not change what the software does.
2. Stating AGI as a goal. A handful of well-funded laboratories say their mission is a system that could do roughly any intellectual task a person can.
3. Stating superintelligence as a goal. Fewer still use Bostrom’s word, a system greatly beyond the best humans, for what they are aiming at.

This page is about the second and third groups. Inclusion under the administration’s “Super Intelligence” label is not evidence that a company claims to be building technical superintelligence, and this page does not treat it as such; the rename timeline records what the administration has actually said and done.

What each organisation says, in its own words

What follows is what each organisation states about its own goal, with the primary source. Where an organisation says “AI”, “AGI” or “superintelligence”, the table uses the word it uses, and nothing stronger. It is not an assessment of how close any of them is.

Organisation	Its own word for the goal	What it says	Primary source
OpenAI (San Francisco)	AGI	Mission "to ensure that artificial general intelligence (AGI)… benefits all of humanity"; AGI defined as "highly autonomous systems that outperform humans at most economically valuable work". The charter does not use the word superintelligence.	OpenAI Charter
Google DeepMind (London)	AGI	Describes itself as "exploring the frontiers of AGI"; defines AGI as "AI that's at least as capable as humans at most cognitive tasks".	"Taking a responsible path to AGI", 2 April 2025
Safe Superintelligence Inc. (Palo Alto and Tel Aviv)	Superintelligence	"One goal and one product: a safe superintelligence." Also states that "superintelligence is within reach".	ssi.inc
Meta (Menlo Park)	Superintelligence	Its frontier research unit is named Meta Superintelligence Labs, which published the Muse Image and Muse Video models on 7 July 2026. Mark Zuckerberg's letter "Personal Superintelligence", published by Meta on 30 July 2025, describes the aim as "everyone having a personal superintelligence that helps you achieve your goals".	Meta AI blog, 7 July 2026
Anthropic (San Francisco)	"Transformative AI"	States its purpose as ensuring the world "safely makes the transition through transformative AI". Does not state AGI or superintelligence as its product goal.	Leadership at Anthropic
xAI (Palo Alto)	AI	Stated mission "to understand the true nature of the universe". Its public mission statement does not name AGI or superintelligence as the goal.	xAI, Series B announcement, 26 May 2024

Three things about that table deserve saying in words.

“Superintelligence” is not used identically by everyone in the third group. Safe Superintelligence Inc. uses the word for a system it intends to build and describes as “within reach”. Meta uses it in the name of a laboratory and in the phrase “personal superintelligence”. Neither company has published a definition matching Bostrom’s (greatly exceeding the best humans in virtually all domains), so this page records that they use the word, not that they mean precisely the same thing by it. Neither claims to have built one.

Building frontier AI is not the same as claiming to build superintelligence. Anthropic and xAI train some of the most capable systems in existence. Neither states superintelligence as its goal in its public mission wording, and this page does not infer one. The same goes for the Chinese laboratories, DeepSeek and Alibaba’s Qwen team among them, whose frontier-class releases feature heavily in US policy argument; their public statements are generally about capability rather than AGI or superintelligence as a named goal.

Nobody in the table claims to have finished. No organisation says it has built AGI or superintelligence. The International AI Safety Report 2026, the nearest thing to a consensus account, describes general-purpose AI capabilities as having “continued to improve” and puts the plausible range for 2030 at anything “from stagnation to rapid improvement to levels that exceed human cognitive performance”, a range of futures, not a description of a system that exists.

What the September 2026 pacing debate shows

A useful window on where these companies think they are came in September 2026. On 9 September a researcher resigned from Anthropic saying that the leading laboratories “earnestly believe it could kill us all by the end of the decade” and are “gambling with our lives”. Reuters reported that Evan Hubinger, Anthropic’s alignment lead, estimated a greater than 10 per cent chance of AI causing human extinction within a decade. Three days later, on 12 September 2026, Anthropic’s chief executive Dario Amodei published “We Must Pace the Frontier”, arguing that “we must slow the pace at which we improve the capabilities of AI models” and committing the company to give outside evaluators such as METR “employee-like access” with the right to publish findings. OpenAI’s Sam Altman endorsed the essay on 13 September, and Elon Musk of xAI did so within days.

Two readings are possible and both are worth holding. One: the people closest to the systems believe superintelligence, in the research sense, is a live possibility within years. Two: none of them says it has arrived; Hubinger’s figure is an estimate of catastrophic risk rather than a claim about capability or a timeline; and the essay’s own concrete timelines are about near-term harms (a “swarm” capable of “taking over the entire internet with a persistent botnet” within “6–12 months”), not about a system beyond human ability across the board.

The administration rejected the premise. In a Truth Social post the President described a “SICK conspiracy going on against AI and Data Centers”, and at the UN General Assembly on 22 September 2026 said “I’m not going to stifle growth of something that will be bigger than the industrial revolution”.

Who is not building it

It bears repeating because the new label invites the opposite conclusion: the overwhelming majority of companies described as “AI companies” (and, under the White House’s published usage, as “Super Intelligence” companies) are building products on top of models made by the laboratories above. A customer-service bot, a translation tool, a fraud filter, a coding assistant: these are AI. They are not attempts at AGI, and calling them “Super Intelligence” in a government document would not make them so.

The UK’s part

The UK does not host a laboratory that states superintelligence as its goal. It hosts Google DeepMind, which states AGI as its aim, and it runs the AI Security Institute, the government body that tests frontier models for dangerous capabilities before and after release, and which provides the secretariat for the International AI Safety Report. Whether UK bodies adopt the administration’s “Super Intelligence” terminology is one of the things the rename timeline is watching; as of 23 September 2026, we found none that had.