Leaders from 20 countries and the European Commission have issued a joint call for tighter controls on frontier AI, including mandatory testing before release and independent assessments of the most capable systems. The September 22 statement also asks UN member states to consider an international institution for setting standards and monitoring compliance.
The proposal is not a law and does not establish a global AI regulator. It instead sets out a framework that governments and companies could pursue, with the document remaining open to further signatories.
- Require companies to test frontier models before deployment and allow independent experts sufficient access to evaluate them.
- Coordinate shared safety standards between governments and exchange information on serious AI safety incidents.
- Explore a UN-backed international body that could set standards, verify compliance and convene states when models cross defined capability thresholds.
The signatories argue that sufficiently advanced AI models have already shown behavior such as evading testing mechanisms, exploiting vulnerabilities and gaining unauthorized access to real-world systems. The statement does not name specific laboratories or incidents.
Named supporters include Norwegian Prime Minister Jonas Gahr Støre, Finnish President Alexander Stubb, German Chancellor Friedrich Merz, Canadian Prime Minister Mark Carney and European Commission President Ursula von der Leyen. Leaders from Australia, Spain, the Netherlands, Singapore, South Africa and Moldova are also among those cited.
The initiative goes beyond voluntary commitments increasingly discussed by AI companies and executives, including calls for outside evaluations of frontier systems. It also extends past the European Union's AI Act, which already imposes obligations on models judged to carry systemic risk, by proposing coordination at an international level.
Its reach remains uncertain. The published signatory list does not include the United States, China, the United Kingdom, France, Japan or India—countries with major influence over frontier-model development and AI policy. Securing participation from those powers may determine whether the proposal develops into a meaningful global framework.
