A lab officially declares it has built superintelligence

Where things stand (11 Oct 2026)

As far as I can find, no covered developer has published a statement that qualifies. The nearest cases are all from Meta, and each reads as an aim rather than a claim:

• Muse launch post (8 Sep, before the question opened): "Muse is a first step" toward personal superintelligence.
• Connect 2026 newsroom post (24 Sep): "with access to Muse, we're building personal superintelligence into devices… something to look forward to."
• Muse for Small Business (29 Sep): "in line with our vision to put superintelligence in the hands of as many people as possible."
• Enterprise Platform post (28 Sep): "We believe superintelligence will create significant new opportunities."

At Connect, Zuckerberg said personal superintelligence is "now within reach" and that "in the coming years… Muse is going to grow into the personal superintelligence." The question's exclusions rule out exactly that kind of language. I give about 2.5% to a lenient resolver counting the 24 or 29 September posts anyway, and I put that mass in the first quarter.

What has changed recently

Trump's renaming order. On 29 September Trump signed Executive Order 14434, which makes "Super Intelligence" and "SI" the executive branch's terms for AI. He has since called anyone who does not adopt them "THE ENEMY". Musk renamed SpaceXAI (formerly xAI) to SpaceXSI, and its X account now reads "SpaceX Super Intelligence". Musk also posted "SpaceX is a super intelligence company." OpenAI says it will not rename itself. Most companies have held off, and one lobbyist expects them to use the term "when communicating with the administration." This raises the chance that a company describes its own products as "super intelligence". But most such uses will be the two-word form or "SI".

How I read the wording rule. I take the literal reading: "Super Intelligence" (two words) and "SI" are not the exact word, and mission or name changes do not count. I give roughly a 35% chance that a resolver would accept the two-word form where it plainly classifies a system.

Capabilities. GPT-6 Astra was released on 3 September. On 8 September an internal OpenAI model solved the Navier–Stokes Millennium Prize problem. Both OpenAI and Anthropic now say AI is beginning to speed up its own development. Even so, OpenAI's official vocabulary stays with "AGI": "personal AGI" and an "automated AI researcher by March 2028". Brockman's "welcome to the AGI era" was said at a press briefing, not written in an official document.

Pressure the other way. Anthropic and OpenAI are pushing to slow ("pace") frontier development. The Ban Artificial Superintelligence Act (Sanders and Casar) now has added House co-sponsors, and a September poll found 68% support for a slowdown and ban. The FTC is investigating OpenAI, Anthropic and others. All of this makes a strong-sense claim to have built superintelligence costly.

Reference class

The comparison is formal milestone declarations by frontier labs. Even at AGI-like capability, none has made an official AGI declaration in its institutional voice. Polymarket puts an OpenAI AGI announcement before 2027 at about 15%. Superintelligence is a higher and more politically charged bar.

Paths to YES, roughly in order

1. Meta declares Muse or a successor "personal superintelligence". This is its stated product path over "the coming years", and its newsroom wording is already edging toward it. I put this at about 25% over five years, concentrated in 2027–29.
2. SpaceXSI describes Grok as a superintelligence in an official post or on the company's X account (X is now owned by SpaceX), either in Trump's generic sense or in the strong sense around Grok 5 or 6. I discount this heavily for the two-word problem: about 10%.
3. Government-facing documents that use the term generically for a company's products.
4. OpenAI, Alibaba or other Chinese labs making strong-sense claims once systems are clearly superhuman: about 6%, 5% and 4% respectively. Google DeepMind and Anthropic are very unlikely.

Allowing for correlation between these, I get about 40% over five years.

Hazard rates

• Q4 2026: about 6.4% (the borderline-post risk plus about 4% for new statements).
• 2027–28: about 2.7–3.0% per quarter, roughly 10.5% a year, driven by the renaming push, Grok 5, Muse's growth and capability gains.
• 2029–31: about 2.4% falling to 1.9% per quarter. The renaming may be reversed after January 2029, and the chance of restrictions on superintelligence rises.

Cumulative: 16.5% by end-2027, 25% by end-2028, 32% by end-2029, 42% by end-2031.

Strongest case against my numbers

• Lower: labs have consistently avoided official milestone claims. Meta may treat "personal superintelligence" as a permanent aspiration, the way it treated "the metaverse". The wording rules are strict. That view suggests 20–25%.
• Higher: capability is accelerating sharply, the US government has made "super intelligence" a generic, politically favoured term, there are ten labs, and the window is more than five years. That view suggests about 55%.

Sources

• Executive Order 14434: whitehouse.gov/presidential-actions/2026/09/inaugurating-the-era-of-super-intelligence/ (29 Sep 2026)
• Meta newsroom: about.fb.com/news/2026/09/the-biggest-news-from-connect-2026/, …/introducing-muse-small-business/, …/launching-meta-enterprise-platform/, …/introducing-muse-personal-ai-agent/
• Moneycontrol and Indian Express on the Connect keynote (24 and 28 Sep)
• OpenAI: openai.com/index/navier-stokes-solution/, …/built-to-benefit-everyone-our-plan/, …/research-acceleration-view-inside-openai/, …/ai-progress-and-recommendations/
• The Verge (10 Oct), TeslaNorth (10 Oct), Washington Examiner, AFP/France24 (9 Oct), PolitiFact (7 Oct)
• Alibaba keynote: alibabagroup.com/en-US/document-2039431633571938304 (22 Sep)
• Polymarket OpenAI AGI market
• darioamodei.com/post/we-must-pace-the-frontier

Question. By the last day of each quarter, has a covered developer officially declared, using the word “superintelligence,” that an AI system it has built is a superintelligence?

Definitions.

• Covered developer: as defined by the site convention.
• Officially declare: an affirmative statement made in the developer’s institutional voice in a developer-authored announcement, report, system card, regulatory or court filing, prepared testimony, or other official document, or a statement published on its own domain. An executive’s signed statement published by the developer counts. Sworn testimony counts only if the qualifying declaration is also published in a developer-authored document or on the developer’s own domain.
• Using the word “superintelligence”: the declaration must contain that exact English word, case-insensitively, as a description or classification of the system. Modifiers such as “artificial” or “personal” are allowed. “Superintelligent,” “ASI,” “AGI,” “superhuman,” and non-English equivalents alone do not count; an official English version published by the developer can qualify on its own publication date.
• A system it has built: an existing AI model or integrated AI system that the declaring developer says it developed or co-developed. It may be internal, unreleased, experimental, or deployed. It need not be named, but the declaration must unambiguously refer to an existing system rather than a future system, proposed design, research programme, or organization.
• Is a superintelligence: the developer affirmatively classifies the system itself as superintelligence. No independent demonstration, benchmark threshold, or particular definition of superintelligence is required. The declaration, rather than the truth of the capability claim, is being measured.

Resolves YES if a qualifying official publication states that an existing system built by that developer is superintelligence, or that the developer has built or achieved superintelligence with an existing system. “We believe our system X is a superintelligence” counts; “X might be a superintelligence” does not. A categorical product announcement can count without technical substantiation if it explicitly makes the classification.

Exclusions (resolve NO).

• Goals, forecasts, plans, claims that superintelligence is “in sight,” “glimpses” or “early signs” of superintelligence, and statements that a system approaches or could become superintelligence.
• Naming a lab, team, product, or project “Superintelligence” without affirmatively classifying an existing system as one. Meta’s July 30, 2025 Personal Superintelligence letter is a statement of goals, not a qualifying declaration.
• Claims only of superintelligence-level performance on a task, or of components or features associated with superintelligence, without classifying the system itself as superintelligence.
• A system’s self-description, third-party characterizations, quotations the developer does not endorse, executive interviews, employee social-media posts, leaks, and anonymous-source reporting.
• No qualifying declaration published by the relevant quarter-end.

Resolving source. Developer-authored official publications and statements on the covered developer’s own domain only, using the developer-source branch of Public confirmation (A2); third-party reporting cannot substitute for the declaration. Apply the Resolution window and Resolution date (A3) conventions. A later withdrawal or dispute does not undo a qualifying declaration.