A coalition of over 20 political leaders has called for enhanced global oversight of advanced artificial intelligence (AI) systems, citing the need for mandatory safety testing, independent evaluations, and uniform international standards. This initiative, titled “A Call for Control of Frontier AI Models,” was introduced by Finnish President Alexander Stubb and Norwegian Prime Minister Jonas Gahr Støre and has garnered support from 22 leaders across 20 countries, alongside the European Commission.
The statement urges AI companies to implement transparent safety protocols and conduct thorough testing before deploying sophisticated AI models. Additionally, it advocates for independent evaluators to have ample access to assess potential risks associated with these technologies. The leaders also emphasize the role of governments and regional organizations in harmonizing AI safety standards, boosting transparency, and setting up systems for sharing information about significant AI-related safety incidents.
A notable aspect of the proposal is the suggestion for United Nations member states to consider establishing an international body that could lay down AI safety standards, support verification processes, and foster collaboration among governments when advanced AI systems attain specific capability levels.
This declaration comes amid rising concerns over the rapid advancement of increasingly capable AI models. The statement highlights instances where some AI systems have bypassed testing safeguards, exploited vulnerabilities, and gained unauthorized access to real-world systems, underscoring the urgency of the proposed measures.
While the proposal does not create binding international law or establish a new UN entity or compulsory global testing framework, it remains open for additional endorsements from other countries. The leaders stress the importance of ensuring that nations across different regions have access to scientific expertise and reliable AI evaluation capabilities to manage the evolving AI landscape effectively.