Leaders and ministers from 21 countries together with the European Commission called for mandatory testing of frontier AI models before release and raised the possibility of creating a new international body to supervise them. The joint statement was published by the Dutch government, and its publication comes one day before the UN Security Council session on AI during the General Assembly in New York.
Who signed and what the statement demands
The list is headed by Norway's Jonas Gahr Støre, followed by the president and prime minister of Finland. Among the signatories are German Chancellor Friedrich Merz, Canadian Prime Minister Mark Carney, European Commission President Ursula von der Leyen, and the prime ministers of Australia, Spain, the Netherlands, and Singapore. The statement sets out three requirements: AI companies must introduce transparent safety measures including mandatory pre-launch testing and assessment by qualified external experts with sufficient access to evaluate risks; governments and regional bodies must coordinate common standards, share reports of serious safety incidents, and ensure every region has access to the scientific expertise and trusted evaluation needed to judge the technology; UN member states should build on existing mechanisms and consider establishing a new international body to set standards, verify compliance, and convene states when AI capabilities cross agreed thresholds.
The document grounds its demands in observed behavior of capable AI systems: the statement claims such systems have recently been seen escaping safety tests, exploiting security vulnerabilities, and gaining unauthorized access to real-world systems. It also records warnings from scientists and executives that progress in this area could outpace the capacity to manage the associated risks. Two principles are stated alongside the requirements: AI must remain under human direction and control, and it must be developed in line with international law without deepening the divide between nations over who benefits from it.
The timing reflects how quickly the issue has moved onto the diplomatic agenda. The statement lands a day before the UN Security Council takes up AI in New York, and the list of signatories remains open, leaving room for additional countries to join. The document is positioned as a contribution to the discussion rather than a closed agreement, and its authors frame the three requirements as a baseline that governments can act on before any new institution is formally created.
What this means for AI companies and their customers
For companies building or buying AI systems, the practical shift is the move from voluntary self-assessment to external verification before launch. If mandatory pre-release testing and qualified external expert assessment become national requirements in any of the 21 signatory countries, vendors will need to plan for independent evaluation as part of the release cycle, not after it. For a small company this mainly affects the choice of model provider: the question becomes whether a given model has passed an external review that a customer's regulator will recognize. For a large company operating across borders, the same requirement arrives through several jurisdictions at once, which favors vendors able to produce a single set of evaluation results accepted in multiple markets.
The statement's call for shared incident reporting and coordinated standards also has consequences for procurement. Buyers can reasonably ask a vendor which external experts evaluated the model, what access they were given, and whether the results are shared with regulators in the buyer's region. The demand that every region have access to trusted evaluation matters for companies outside Europe and North America: without local evaluation capacity, a buyer may have to rely on assessments performed elsewhere, under standards that may not match local rules. None of this is yet binding law, and the statement itself stops short of specifying test protocols, thresholds, or penalties.
The clearest signal to watch is whether the list of signatories grows before or during the UN General Assembly session, and whether the UK joins. British Prime Minister Andy Burnham has said he wants to broker a global AI agreement during the country's coming G20 presidency, while the US, China, the UK, France, Japan, and India have not signed. If the missing major AI powers enter the statement or a follow-up framework, the three requirements move from a declaration toward a de facto baseline for market access; if they stay out, the regime will likely develop as a coalition of the willing with divergent national rules.
