Leaders from more than 20 countries urged on Monday to adopt binding security measures for advanced artificial intelligence (AI), warning that the development of this technology could advance faster than the ability to control it.
In a joint statement, the signatories pointed out that border AI models pose “serious security risks” if mismanaged.
The group cited recent cases of AI programs that circumvented test safeguards and obtained unauthorized access to real-world systems, such as in the widely advertised Hugging Face intrusion.
In July, two OpenAI models left an isolated test environment and autonomously attacked Hugging Face, an open-source platform.
The signatories include the leaders of Norway, Finland, Australia, Canada, Germany, Spain, Ireland, the Netherlands, Singapore, South Africa and Kenya, as well as the President of the European Commission, Ursula von der Leyen.
The AI superpowers, the United States and China, as well as the United Kingdom, France, Japan and India, are not part of the group.
The statement remains open to membership of other members.
The signatories called on companies to adopt transparent security protocols, including mandatory testing and independent assessments.
AI “must remain under human direction, supervision and control,” the leaders said, adding that stricter supervision should not widen the gap between countries in access to AI benefits.
The statement comes ahead of the UN General Assembly in New York, where the growing role of AI in society and its threat to employment and cybersecurity will take a central place on the agenda.
The leaders are expected to discuss an international framework there.
U.S. President Donald Trump has strongly opposed security measures on AI and promised on Saturday not to allow the “destruction” of this industry.
