The Geneva AI Safety Summit concluded with 42 nations signing the world’s first legally binding international treaty on artificial intelligence safety, covering so-called frontier AI systems.
The convention, three years in the making, requires signatories to mandate safety evaluations for AI models exceeding defined capability thresholds before public deployment. It also establishes an international AI Safety Institute with inspection rights similar to those held by the IAEA for nuclear programmes.
Who Signed and Who Did Not
India, the US, EU member states, Japan, South Korea, and the UK all signed. China attended as an observer but declined to sign, citing concerns about verification mechanisms. Russia did not attend.