Twenty-plus nations and the EU call for frontier AI controls

Leaders from more than twenty countries plus the European Commission signed a joint call — hosted on the Finnish Presidency site — for tighter control of frontier AI models: mandatory company pre-deployment testing with independent evaluators, shared government incident reporting, and exploration of a UN-linked institution that could set standards, verify compliance, and convene states when capability thresholds are crossed. The United States and China are not on the posted signatory lists.

The declaration treats frontier models as systems that have already circumvented testing safeguards, exploited vulnerabilities, and gained unauthorized access to real-world systems — and says AI must stay under human direction, oversight, and control, in line with international law.

The rule in English

Core claim. Rapid frontier-AI development poses serious safety and security risks if unmanaged. Signatories welcome existing risk-management initiatives and add three asks.

Ask (1) — companies. Transparent safety protocols, including mandatory pre-deployment testing and independent evaluation, with qualified evaluators granted sufficient access to assess risks.

Ask (2) — governments and regional organisations. Coordinate common standards; strengthen transparency, including shared reporting of serious safety incidents; ensure countries in all regions get scientific capacity, expertise, and trusted evaluation — without widening the benefit gap between countries.

Ask (3) — UN member states. Build on existing international mechanisms and explore creating an international institution able to set standards, enable verification, and convene states when capability thresholds are crossed.

Roster (primary). Finnish page lists heads of government/state and foreign ministers from Australia, Austria, Bahrain, Canada, Denmark, Estonia, Finland, France, Germany, Iceland, Ireland, Kazakhstan, Kenya, Latvia, Liechtenstein, Luxembourg, Moldova, Netherlands, Norway, Romania, Sierra Leone, Singapore, South Africa, Spain, Türkiye, and the UAE, plus European Commission President Ursula von der Leyen. Finland appears twice (President Stubb and Prime Minister Orpo). Late endorsements noted on the page: Austria and Romania (Sept. 21); Liechtenstein and Luxembourg (Sept. 22); France and Sierra Leone (Sept. 23). Statement remains open for more leaders. The Sep 22 Netherlands government mirror shows the earlier 20-nation + EU core before those late adds.

Absent. No U.S. or China signatory on either posted list.

Who wrote it / who it binds

Writers: joint leaders (Finland/Norway hosts prominent). A nonbinding diplomatic declaration. Binds no legislature and creates no new body by itself — Ask 3 is exploration, not a charter.

Who it exempts

Non-signatories (including the U.S. and China). Open- and closed-model developers are not named as exempt from Ask 1; the text points at “companies” generally. National legal incorporation of any future standards is left to governments.

What a founder, parent, or worker should do differently

Frontier founders: treat mandatory pre-deployment testing, independent evaluator access, and serious-incident sharing as the bar this bloc wants written into coordinated standards — even if Washington and Beijing stay off the letter.

Policymakers: this is a multilateral instrument after Altman’s Sept. 23 UNSC speech — same human-control theme, different authors (states + EU, not a lab CEO). Watch whether Ask 3 becomes a real institutional proposal or stays a press-line.

Parents and workers: not a school or consumer-app rule — diplomatic governance of frontier model testing and verification.

What’s next

• Whether more G20/major-power leaders endorse — or Washington/Beijing answer with a competing track.
• Concrete next step on the “international institution” (TORs, host, verification powers) vs. speech-only.
• Company reaction to mandatory pre-deployment testing + evaluator access (ties to prior embedded-evaluator / CAISI coverage).

Source: Office of the President of Finland, “A Call for Control of Frontier AI Models,” Sept. 2026.
Mirror: Government of the Netherlands, same title, Sept. 22, 2026.