Superintelligence statement vs other AI risk statements

The statements side by side

Signature counts are snapshots: most of these pages keep collecting names. “Seen” is the date we read the number. Each statement’s name links to the original text.

Where the sources disagree

• -Superintelligence statement wording. The signing page reads “...not lifted before there is broad scientific consensus... and strong public buy-in.”4 FLI’s press release quotes the “full statement” with numbering: “...not lifted before there is 1) broad scientific consensus that it will be done safely and controllably, and 2) strong public buy-in.”14 The words are the same apart from the “1)” and “2)”.
• -Superintelligence statement date. FLI’s newsletter says “released October 22”, and CyberScoop reported the launch on 22 October 2025.1213 FLI’s press release page has a published date of 27 March 2026 in its metadata.14 We use 22 October 2025.
• -Superintelligence statement count. In one FLI newsletter the headline says “Over 65,000” and the text says “almost 65,000”.12 The live counter is now higher (77,591, seen 2026-10-15) and includes 5,000 signatures added from an Ekō petition.4
• -Pause letter date. FLI’s page says “Published 22 March, 2023”.3 Other sources give 28 or 29 March.
• -Paris signatories. Launch-day reporting said “60 countries”,7 while the Élysée page lists 65 entries (seen 2026-10-15): 63 countries and two organisations.6 The page doesn’t say whether any were added later, so we show both numbers.

How the asks escalated, and how they differ

Reading the statements in order, the asks change in kind, not only in strength.

1. 1.“Make it a priority” (2023). The CAIS sentence names extinction risk but asks only that reducing it be “a global priority”.2 The Will AI kill us all? guide covers who signed, including the heads of OpenAI, Google DeepMind and Anthropic. The statement gives no probability, no timeline and no policy.
2. 2.“Stop for six months” (2023). The FLI letter asked for a time-limited pause on training “more powerful than GPT-4”. It said governments “should step in and institute a moratorium” if labs did not pause.3
3. 3.“Draw lines no system may cross” (2024–2025). The IDAIS-Beijing scientists listed specific red lines, for example “No AI system should be able to copy or improve itself without explicit human approval and assistance.”9 The 2025 Global Call for AI Red Lines asks governments to turn red lines into “an international agreement” with “robust enforcement mechanisms — by the end of 2026.”11
4. 4.“Don’t build it at all, for now” (2025). The Statement on Superintelligence moves from a pause on bigger models to a “prohibition” on one goal, superintelligence, until there is “broad scientific consensus” and “strong public buy-in”.4

Two statements sit to one side of that line:

• -Seoul puts the stopping rule inside the companies. Each firm sets its own risk thresholds, and “in the extreme” commits not to develop or deploy a model it cannot keep below them.1 The commitments describe themselves as “voluntary”.1
• -A Right to Warn asks for nothing about AI models. It asks about the people who work on them: companies should not “enforce any agreement that prohibits ‘disparagement’ or criticism of the company for risk-related concerns”, and should not retaliate against staff who go public “after other processes have failed”.10 Its stated reason: “AI companies have strong financial incentives to avoid effective oversight”.10

The superintelligence statement (October 2025): what it says and who signed

The Statement on Superintelligence was launched by the Future of Life Institute on 22 October 2025.1213 Its ask: “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.”4 It covers superintelligence only, not AI in general, and the ban it asks for lifts once there is scientific consensus and public buy-in.4

Who signed: Yoshua Bengio, Geoffrey Hinton and Stuart Russell are named as signers.12 For the September 2026 bill to ban superintelligence in the US Congress, see our explainer on the bill.

What governments signed vs what scientists signed

The two government declarations are weaker than what scientists and staff signed, and they got weaker over time.

• -Bletchley (2023) accepts the “catastrophic” framing and says that those building frontier AI “have a particularly strong responsibility for ensuring the safety of these AI systems”.5 Its commitments are to research and cooperation: “an internationally inclusive network of scientific research on frontier AI safety” and “building respective risk-based policies across our countries”.5 It doesn’t mention extinction, existential risk, a pause or a moratorium (we searched the full text). Its 29 original signatories include both the US and China.5
• -Paris (2025) focuses on access and inclusion. Among its listed priorities are “Promoting AI accessibility to reduce digital divides” and “avoiding market concentration”.6 On safety, it says it will “keep addressing the risks of AI to information integrity”. It “note[s] the voluntary commitments” made at Seoul.6 The words “catastrophic”, “existential”, “extinction”, “frontier” and “severe” do not appear in it. China signed it.67 The US and UK did not. A UK government spokesperson said the declaration “didn’t provide enough practical clarity on global governance, nor sufficiently address harder questions around national security and the challenge AI poses to it.”7 The Guardian reported the snub alongside US vice-president JD Vance’s criticism of Europe’s “excessive regulation”.7
• -Seoul (2024) is a hybrid. Two governments announced it, but companies signed it, among them Amazon, Anthropic, Google, Meta, Microsoft, OpenAI, xAI and the Chinese firm Zhipu.ai.1

The scientist and public statements (CAIS, the FLI letters, IDAIS, the Red Lines call) use stronger risk language and make more specific asks, but bind no one. The same few names recur across them. Yoshua Bengio, Geoffrey Hinton and Stuart Russell, for example, endorsed A Right to Warn,10 appear among the IDAIS-Beijing signatories,9 and are named as signers of the Statement on Superintelligence.12

What each one led to (only where a source says so)

We list a result only when a source we opened states it. “Not shown” means we found no such source, not that nothing happened.

• -FLI pause letter: no lab announced a pause in any source we found.
• -CAIS statement: not shown.
• -Bletchley Declaration: it ended “We look forward to meeting again in 2024”.5 The Seoul summit commitments were published on 21 May 2024,1 and the Paris statement says the Bletchley and Seoul summits “have been essential in progressing international cooperation on AI safety”.6 New Zealand joined the declaration in October 2024.5
• -IDAIS-Beijing: the statement says “At future International Dialogues we will build on this list”.9 The IDAIS site lists other dialogues (London, Shanghai, Venice, Oxford).9 We did not open them for this page.
• -Seoul commitments: the companies undertook to publish “a safety framework focused on severe risks by the upcoming AI Summit in France”.1 Whether each one did is not covered here.
• -A Right to Warn: not shown.
• -Paris statement: the US and UK declined to sign.7
• -Global Call for AI Red Lines: its own deadline is “by the end of 2026”.11 Its FAQ suggests states “could use the AI Impact Summit in India in February 2026 to endorse initial red lines”.11 That is a proposal on the campaign’s page, not a record of what happened.
• -Statement on Superintelligence: signatures grew from “more than 700” at launch13 to 77,591 (seen 2026-10-15).4 In September 2026, a bill to ban superintelligence was introduced in the US Congress. Our sources do not say the statement caused it. See our explainer on the bill.

Common misquotes to watch for

• -“Experts say AI will kill us.” No statement here says that. The CAIS sentence calls extinction a risk to prioritise. A Right to Warn says loss of control could potentially result in extinction.210
• -“World governments agreed AI is an extinction risk.” Bletchley says “catastrophic”, not extinction. Paris drops catastrophic-risk language altogether.56
• -“The superintelligence statement wants to ban AI.” It covers superintelligence only, and the ban lifts once there is scientific consensus and public buy-in.4
• -“The pause letter wanted to stop AI.” The letter itself says: “This does not mean a pause on AI development in general”.3