The Statement on Superintelligence, Explained: Inside the Prohibition Call

The Statement on Superintelligence is a one-sentence public petition, hosted at superintelligence-statement.org and identified on its own page as “Hosted by Future of Life Institute.” It reads, in full:

“We call for a prohibition on the development of superintelligence, not lifted before there is broad scientific consensus that it will be done safely and controllably, and strong public buy-in.”

As of this writing the statement carries 74,781 signatures (verified directly against the live page; the count is real-time and will have moved by the time you read this). This guide explains what the statement actually asks for, who has put a verified name to it, and — the part most explainers skip — how a blanket prohibition proposal like this one relates to, and differs from, the model-specific “threat model” vocabulary that individual frontier AI labs use in their own safety frameworks.

What the statement calls for, precisely

Read the sentence closely, because the three clauses are doing distinct work:

• “A prohibition on the development of superintelligence” — not a pause on deployment, not a licensing regime, not a compute cap set at a specific FLOP threshold. The ask is a stop on development itself, aimed at a capability class (superintelligence) rather than any named model, lab, or jurisdiction.
• “Not lifted before there is broad scientific consensus that it will be done safely and controllably” — the prohibition has an explicit release condition, but that condition is a scientific-consensus threshold, not a calendar date or a specific technical benchmark.
• “and strong public buy-in” — a second, independent condition. Scientific consensus alone does not lift the prohibition under the statement’s own terms; public support is named as a separate, additional bar.

This makes the statement a moratorium-until-consensus proposal, a category distinct from the graduated, threshold-based approach used in the safety frameworks CASRAI covers elsewhere on this site (Responsible Scaling Policies, Preparedness Frameworks, Frontier Safety Frameworks — see What Is a Frontier AI Framework?). Those frameworks define capability thresholds and pre-committed mitigations a lab applies to its own models as it develops them. The statement instead asks that development of an entire capability class not proceed at all until two external conditions are met. It is a policy demand aimed at the field, not an operating document any single organization has adopted for itself.

Who is behind it, and who has actually signed

The statement is a public, open-signature petition rather than an institutional policy release. Anyone can add a name, which is exactly why the signature count needs to be read carefully rather than treated as a roster of vetted experts. Two things are worth separating:

First, the page itself: it is hosted by the Future of Life Institute, the nonprofit also behind the 2023 “Pause Giant AI Experiments” open letter, but hosting is not the same claim as sole authorship, and the page does not present it as a Future of Life Institute policy document — it presents it as a statement other people and organizations can sign onto.

Second, the signatories themselves. Checked directly against the statement’s own published signatory data rather than secondary reporting (a French outlet has separately named him as a signatory, but that single secondary mention was not treated as sufficient on its own): Yoshua Bengio appears as a verified signatory, listed with a confirmed name/email match and the site’s highest verification badge tier, credited as a Turing Award laureate and among the most-cited computer scientists alive. Geoffrey Hinton appears the same way, listed as a Nobel and Turing laureate. Both are figures who have previously and separately spoken about catastrophic AI risk, which is relevant context for why their names carry weight on a statement like this one — but their presence on the list is a fact about this specific petition, verified against the petition’s own data, not an endorsement CASRAI is asserting on their behalf beyond what the statement’s own page shows.

Because the list is open-signature, treat the raw count as a measure of public reach and mobilization, not as a poll of AI safety researchers or a scientific-consensus vote — the statement’s own second clause explicitly distinguishes “broad scientific consensus” from the act of signing, which is one reason the two shouldn’t be conflated when citing this page.

Why this is showing up in more searches and coverage now

“Superintelligence” as a search term has not spiked in a single viral moment; it has climbed steadily. Citation volume for the term across tracked news coverage rose roughly 4.5x between January 2025 and October 2025, increasing month over month rather than in one burst — the pattern of a topic entering sustained public conversation, not a one-week news cycle. Google Trends search-interest data for “superintelligence” shows a more modest but real rise over the same broader window, with a step up from a low-double-digit baseline to a sustained mid-teens-to-twenties level, plus a smaller uptick around mid-September that plausibly tracks the same existential-risk news cluster this statement sits inside — supporting context for the trend, not the headline claim on its own. Set against that backdrop, a fast-growing, laureate-signed petition calling for an outright development prohibition is a natural amplifier: it gives a concrete, quotable artifact to a conversation that was already building.

The NIKOLAI angle: a prohibition proposal operates at a different level than a lab’s own threat-model vocabulary

CASRAI maintains NIKOLAI, CASRAI’s own independent, unendorsed reference dictionary for frontier-AI-safety terminology — not a standard, and not something any lab, evaluator, or regulator has adopted or approved. It exists because labs, governments, and researchers each describe overlapping safety concepts in incompatible vocabularies, and NIKOLAI is an attempt at a shared index for looking a term up, nothing more.

One of NIKOLAI’s ten tracks, N2 — Threat Models and Risk Framing, defines the vocabulary labs use to describe how a specific model could cause catastrophic or extinction-level harm. Two of its elements are directly relevant here:

• Threat model — the specific, named mechanism by which a capability could lead to catastrophic outcomes (for example, uplift toward a biological or cyber weapon), the unit a lab’s Responsible Scaling Policy or Preparedness Framework is actually built around.
• Risk pathway — the chain of steps from a capability existing to that harm materializing, which is what a lab’s mitigations and evaluations are designed to interrupt.

Here is the distinction worth being precise about: NIKOLAI’s N2 vocabulary describes risk within a given model or lab’s own framework — a specific threat model, tied to a specific capability, mitigated by that lab’s specific safeguards, as covered on CASRAI’s What Is a Frontier AI Model? and Frontier AI Labs pages. The Statement on Superintelligence operates one level up: it isn’t proposing a threat model for a particular system, and it isn’t asking any lab to run a specific evaluation or adopt a specific mitigation. It’s a categorical demand that development of an entire capability tier not proceed until two external conditions — scientific consensus and public buy-in — are met, independent of what any individual lab’s own threat-model analysis concludes. A lab could, in principle, judge its own threat models as adequately mitigated under its own framework while the statement’s broader consensus-and-buy-in bar remains unmet; the two operate on different objects (a specific system’s risk pathway versus an entire capability class) and by different logic (technical mitigation versus a public/scientific-consensus gate). Readers coming from CASRAI’s framework-comparison content should not read the statement as a threat model itself, or as a claim NIKOLAI’s dictionary takes a position on — NIKOLAI documents the vocabulary the field uses; it does not adjudicate whether a prohibition proposal like this one is the right policy response.

How this differs from CASRAI’s incident-reporting and framework-comparison content

Most of CASRAI’s frontier-ai-safety content documents how existing frameworks and regulations already function: what a lab’s Responsible Scaling Policy commits to, what SB 53 or the RAISE Act require to be reported, how evaluators like METR fit into the picture. This page is different on purpose: it documents an active advocacy position — a call for something that does not exist yet (a binding development prohibition) — rather than a policy or framework already in force. Nothing about the statement being widely signed makes it binding, and nothing here should be read as CASRAI taking a position on whether the prohibition it calls for is the correct policy. The goal is the same as the rest of this cluster: give an accurate, sourced account of a term and a document people are searching for, so the terminology is legible regardless of where someone lands on the underlying policy question.

Frequently asked questions

Is the Statement on Superintelligence a law or binding policy?

No. It is a public petition/open letter. Signing it carries no legal or regulatory force; it is a mechanism for public and expert opinion to be expressed and counted.

Who runs superintelligence-statement.org?

The page identifies itself as hosted by the Future of Life Institute, the nonprofit that also organized the 2023 “Pause Giant AI Experiments” letter.

Does a large signature count mean scientific consensus exists?

No, and the statement’s own wording treats the two as separate things — it asks for “broad scientific consensus” as one condition and “strong public buy-in” as a second, distinct condition, rather than treating high sign-up numbers as satisfying either on their own.

Is “superintelligence” the same thing as a “frontier AI model” in CASRAI’s other guides?

No. A frontier AI model is a present-tense regulatory and technical category (see What Is a Frontier AI Model?) already in use by developers and regulators today. Superintelligence, as the statement uses the term, refers to a hypothesized future capability tier beyond current systems — the statement is about whether development toward that tier should proceed, not a claim that it has been reached.