"A Call for Control of Frontier AI Models" is a joint statement of heads of state and government initiated by Norwegian Prime Minister Jonas Gahr Støre and Finnish President Alexander Stubb and launched on 21 September 2026. Primary text: A Call for Control of Frontier AI Models (joint statement, September 2026).
Status
A political statement, not a treaty or resolution. It creates no obligations, establishes no body and sets no thresholds. Its operative verbs are "call for," "welcome" and "explore." It remains open for endorsement by leaders of other countries, and its signatory list has changed several times since launch (A Call for Control of Frontier AI Models (joint statement, September 2026)).
Background
The statement was launched on the sidelines of the 81st session of the UN General Assembly in New York (Source: straitstimes.com). The Norwegian government's launch press release gives as background that researchers, experts and heads of the largest US AI companies had warned about the pace of development "after several leading models broke out of their test environments and attacked other organisations without being discovered," and that the UN Secretary-General had described runaway AI as one of three existential threats to humanity, alongside the climate crisis and increasing inequality (Source: regjeringen.no). The incidents referred to are those collected at AI Agent Containment Incidents, 2026; the statement itself cites "capable AI systems circumventing testing safeguards, exploiting vulnerabilities and gaining unauthorized access to real-world systems" without naming them.
The Wall Street Journal reported on 23 September that Støre had recruited Canada by text message, and that Prime Minister Mark Carney replied within ten minutes: "Of course. We're in." The same report says Carney had discussed AI regulation with French President Emmanuel Macron on Saint Pierre and Miquelon as Macron's government asked the UN Security Council to focus on AI and security, and describes the signatories as "a club of Western presidents and prime ministers" whose push for global controls echoes the nuclear arms treaties of the early Atomic Era; only the article's opening was retrieved (Source: wsj.com). France chaired the UN Security Council's open meeting on AI and international security two days after the launch.
In launching it, Støre said the signatories would address the issue further during the General Assembly's high-level week, and Stubb said: "We have a choice: we can compete, cooperate or put on the brakes. To me, cooperation is the best way forward" (Source: regjeringen.no).
Key provisions
The statement makes three calls, each addressed to a different set of actors (A Call for Control of Frontier AI Models (joint statement, September 2026)):
- Companies — transparent safety protocols, "including mandatory pre-deployment testing and independent evaluation, with qualified evaluators granted sufficient access to assess risks."
- Governments and regional organisations — coordinated common standards, greater transparency "including shared reporting of serious safety incidents," and access for countries in all regions to "scientific capacity, expertise and trusted evaluation."
- UN member states — building on existing international mechanisms and exploring "an international institution, able to set standards, enable verification, and convene states when capability thresholds are crossed."
It also states that "AI must remain under human direction, oversight and control," that AI must be developed and used in line with international law, and that oversight should be strengthened without widening the gap between countries in access to AI's benefits. It welcomes "recently launched initiatives" to manage frontier risks without naming them. The Straits Times, reporting the statement, listed as recent initiatives Microsoft's draft code of conduct for its models of 14 September (Humanist AI Code of Conduct for MAI Models (Microsoft AI, first draft, September 2026)), Anthropic's announcement on 18 September of a partnership with Accenture on independent evaluation, and OpenAI's safety and security proposals of 21 September (Building Standards for the Next Phase of AI (OpenAI, September 2026)); that list is the newspaper's, not the statement's (Source: straitstimes.com).
Signatories
| Date | Signatories | Source |
|---|---|---|
| 21 September 2026 (launch) | 22 leaders, per the Norwegian government | (Source: regjeringen.no) |
| 22 September 2026 | Singapore, the EU and 19 other countries, per The Straits Times | (Source: straitstimes.com) |
| Endorsements through 23 September 2026 | 28 leaders from 26 states and the European Commission (earlier PDF capture) | A Call for Control of Frontier AI Models (joint statement, September 2026) |
| Endorsements through 24 September 2026 | 30 leaders from 28 states and the European Commission | (Source: regjeringen.no) |
The 28 states of 24 September are Australia, Austria, Bahrain, Canada, Croatia, Denmark, Estonia, Finland, France, Germany, Iceland, Ireland, Kazakhstan, Kenya, Latvia, Liechtenstein, Luxembourg, Moldova, the Netherlands, Norway, Portugal, Romania, Sierra Leone, Singapore, South Africa, Spain, Türkiye and the United Arab Emirates. Endorsements after launch are dated in the document: Austria and Romania on 21 September, Liechtenstein and Luxembourg on 22 September, Croatia, France and Sierra Leone on 23 September, and Portugal on 24 September. The full list of named leaders is on the source page.
The United States, the United Kingdom and China are not signatories. Among G7 members, Canada, France and Germany signed and the United States, the United Kingdom, Japan and Italy did not. Singapore is the only Southeast Asian signatory (Source: straitstimes.com).
Reception
UN Secretary-General António Guterres welcomed the launch on X on 21 September, describing it as calling on UN member countries "to build on existing international mechanisms & explore creating an international institution able to set standards, enable verification & convene states when capability thresholds are crossed." In the same post he welcomed the first thematic brief of the Independent International Scientific Panel on AI, released that day, and encouraged external experts, including from frontier AI labs and safety institutes, to engage with the Panel (Source: aninews.in).
Singapore's Prime Minister Lawrence Wong explained Singapore's endorsement in a Facebook post on 22 September: "because these risks cross borders, effective safeguards will require international cooperation," adding that "AI holds tremendous promise, but it must always remain under human control and oversight" (Source: straitstimes.com).
The United States government set out the opposite position during the same week. President Donald Trump told the General Assembly on 22 September that the United States "totally rejects any attempt to construct a globalist scheme to control" artificial intelligence (Source: theverge.com), and OSTP Director Michael Kratsios told the Security Council that the rapid advance of the frontier "is not a reason to pause its further development or to constrain it with new global governance structures" (Source: insideaipolicy.com).
Comparison with other approaches
| Instrument | Date | Parties | Addressees | Institution proposed |
|---|---|---|---|---|
| Bletchley Declaration | November 2023 | 28 countries and the EU, including the US and China | States, developers | None; a shared evidence base |
| Paris AI Action Summit Declaration | February 2025 | 64 signatories; the US and UK declined | States | None |
| A Call for Control of Frontier AI Models | September 2026 | 30 leaders from 28 states and the European Commission as of 24 September; the US, UK and China absent | Companies, governments and regional organisations, UN member states | An international institution to be explored, with standard-setting, verification and convening functions |
Relative to the Bletchley Declaration, which assigned frontier developers responsibility for safety testing and evaluations, the statement adds the terms "mandatory" and "independent" and an evaluator-access condition, and it adds a verification function at the international level. Relative to the pacing proposals of September 2026 compared at Coordinated slowdown proposals compared, it asks for neither a pause nor a slowdown. OpenAI's position paper of the same day proposes international technical standards developed through national AI safety institutes and states that such standards "would not be licenses, mandatory prerelease review, or approval requirements" (Building Standards for the Next Phase of AI (OpenAI, September 2026)); the statement's first call uses the word "mandatory" for pre-deployment testing without saying who would require it.
Open questions
- Whether any signatory government proposes a text, mandate or host for the international institution the third call asks states to explore, and through which "existing international mechanisms."
- How "capability thresholds" would be defined for the convening function, and by whom.
- Whether the United States, the United Kingdom, China, Japan or India endorse the statement while it remains open.
Relationships
- instance-of: Frontier AI Governance
- depends-on: A Call for Control of Frontier AI Models (joint statement, September 2026) — primary text.
- related: Jonas Gahr Støre, Alexander Stubb — initiators; António Guterres — welcomed the launch.
- related: Ursula von der Leyen, Anthony Albanese, Lawrence Wong — signatories.
- related: UN Security Council, Independent International Scientific Panel on AI, Global Dialogue on AI Governance — UN bodies active during the launch week.
- related: Bletchley Declaration (AI Safety Summit, 2023), Paris AI Action Summit Declaration (2025) — earlier multilateral declarations.
- related: Coordinated slowdown proposals compared, AI Agent Containment Incidents, 2026.
- related: AI Pre-Release Vetting, Third-Party AI Assessment, AI Incident Reporting, International AI Standards, AI Control.
- contradicts: Donald Trump, Michael Kratsios — the US administration's rejection of new global AI governance structures in the same week.