A group of 20 countries, alongside European Commission President Ursula von der Leyen, signed a declaration on the sidelines of the UN General Assembly High-level Week in New York, calling for frontier AI to remain under human direction, oversight and control. The initiative was spearheaded by Finnish President Alexander Stubb and Norwegian Prime Minister Jonas Gahr Store. The declaration warns that advanced AI systems have already circumvented testing safeguards, exploited vulnerabilities and gained unauthorized access to real-world systems.
The world's two leading AI powers, the US and China, did not sign the declaration, nor did several major European players, including the UK, France and Italy. However, Germany, Spain and a number of smaller EU states joined the initiative, which remains open for additional endorsements. The declaration calls for mandatory pre-deployment testing, independent evaluations with sufficient access to frontier models, common international standards and shared reporting of serious safety incidents.
The initiative stops short of demanding a pause or slowdown in AI development, instead seeking to create an international institution capable of setting standards, verifying compliance and convening governments when AI systems cross certain capability thresholds. Finnish President Alexander Stubb acknowledged that bringing the dominant AI powers on board will be crucial, saying the goal was to create momentum and eventually attract both Washington and Beijing.
Stubb said his preferred model would resemble the International Atomic Energy Agency, although organizers deliberately left the institutional format open to avoid alienating governments wary of creating another UN body. The push comes as Washington and Beijing are increasingly setting the terms of the global AI debate themselves and promoting competing international frameworks.
The US has expanded its Pax Silica initiative to 24 partners, focusing on securing semiconductor, critical-mineral and AI supply chains among trusted partners. China, meanwhile, established the World Artificial Intelligence Cooperation Organization in July with 29 founding countries, including Russia, many from the Global South. China has paired this initiative with open-weight models, training programs and infrastructure projects, presenting itself as an alternative to Western-controlled AI ecosystems.
The UN has also launched its own Global Dialogue on AI Governance, meaning the new European-led initiative enters an increasingly crowded contest over who gets to write global AI rules. Some countries are keeping a foot in multiple emerging AI governance camps rather than choosing a single bloc. Kazakhstan has joined China-backed WAICO and the US-led Pax Silica initiative while also signing the human control declaration.
Kenya and South Africa, founding members of WAICO, have now backed the new European-led initiative as well. The goal is to create momentum and eventually attract both Washington and Beijing to the initiative, with Stubb arguing that humans need to be in charge of AI systems. The initiative's backers want to ensure that frontier AI systems remain under human control, with mandatory testing and evaluation to prevent potential risks.
Key points
- The EU-backed coalition aims to establish a global AI order with human oversight and control.
- The initiative, signed by 20 countries, calls for mandatory pre-deployment testing and independent evaluations of frontier AI systems.
- The US and China, the world's leading AI powers, have not signed the declaration, but the initiative remains open for additional endorsements.
