Norwegian–Finnish appeal for international AI oversight gathers 28 countries — but without the US
A Finnish–Norwegian initiative statement, backed by 28 countries and the European Commission, demands mandatory safety testing of frontier AI models before they are deployed and proposes exploring an international institution for standards and verification. The call gained momentum during the UN General Assembly — but meets direct resistance from Washington, and it remains unclear who has actually signed and what could make the demands binding.
What the call actually demands
The initiative, titled "A Call for Control of Frontier AI Models," was launched by Norway's Prime Minister Jonas Gahr Støre and Finland's President Alexander Stubb. According to MLex, which cites the official statement of 29 September 2026, the call is backed by leaders from 28 countries and the European Commission, and makes concrete demands:
- Mandatory testing before deployment and independent evaluation of frontier AI models — meaning companies do not themselves decide whether and how models are tested.
- Sufficient access for evaluators, so that independent oversight can actually be carried out.
- Coordination between governments on standards and international reporting of safety incidents.
- That UN member states explore an international institution that could "set standards, enable verification, and convene states," as MLex quotes the statement — in The Irish Times' account, when capability thresholds are crossed.
The document's own rationale is that rapid development of frontier AI models poses serious risks if left ungoverned. "AI must remain under human direction, oversight and control. It must be developed and used in accordance with international law," the call states, as reproduced by Reclaim The Net. Among the cited risks is that AI could circumvent the testing regime's safeguards and gain unauthorized access to systems in the real world.
How many have signed? The numbers don't add up
The scale of support is the news value now — but the sources are not in agreement:
- MLex (29 September 2026): leaders from 28 countries plus the European Commission.
- Reclaim The Net (28 September): roughly 30 leaders, including Canada's Mark Carney and France's Emmanuel Macron.
- The Irish Times (28 September): more than 20 leaders, including Ireland's Taoiseach Micheál Martin.
The discrepancy may reflect different points in time, different definitions (countries versus individuals), or different tallies during UN General Assembly week. But it also means the exact signatory list cannot be verified against the primary source in the available evidence — all coverage is secondary reporting on the statement. What is certain is that the call encompasses leaders from dozens of countries, including several large EU states, but neither the US nor China.
The resistance: Washington says "globalist scheme of control"
The geopolitical fault line is clear. According to The Irish Times, President Donald Trump has said that Washington will oppose any "globalist scheme of control" for AI. Science and technology advisor Michael Kratsios elaborated the position to the UN Security Council: the dialogue on AI, "in this forum and others, cannot be allowed to drift toward global governance."
China occupies a third position. The country, which per The Irish Times is the only major AI developer outside the US, has not signed the Norwegian–Finnish call — but unlike Washington, Beijing, according to China's ambassador Fu Cong, is not opposed to international oversight of the technology within the UN framework.
That means there is no united bloc against oversight, but rather two indistinct fronts: a broad group of states demanding mandatory measures, and two great powers disagreeing about where and how — not necessarily whether.
What the call is — and is not
Precision matters here: this is a non-binding call. None of the sources document enforcement mechanisms, deadlines, or a formal signing process. Nor do any of the available sources say which companies would be covered, or how mandatory testing would in practice be enforced across jurisdictions.
One anecdote, relayed by Reclaim The Net with reference to The Wall Street Journal, illustrates the diplomacy's method: Støre is reported to have texted Carney to invite Canada on board, with the reply reported as "Of course. We're in." It is worth noting that this is second-hand reporting, not directly verified.
What real control would take
For any of the demands to become more than words, at least three things must happen: a UN process that formally takes up the institutional proposal; a definition of which models trigger a testing obligation (the capability threshold is mentioned but not specified in the available sources); and at least indirect support from the US or China, since no mandatory regime works without the two largest developers. China's openness to UN oversight and the US's categorical no mean the diplomatic space lies within the UN system — precisely where the US does not want it discussed.
The open questions therefore outnumber the answers: who has actually signed, whether the institutional proposal gains a foothold in any UN process, and how industry would be affected — something the available evidence says nothing about. The call sets the agenda. It does not establish oversight.