The Singapore Consensus on Global AI Safety Research Priorities

Summary

The Singapore Consensus on Global AI Safety Research Priorities (arXiv:2506.20702) is a consensus document produced by the 2025 Singapore Conference on AI (SCAI), held April 26, 2025, alongside ICLR 2025. Authored by Yoshua Bengio, Stuart Russell, Max Tegmark, Dawn Song, and 84 additional co-authors from 11 countries, it builds on the 2025 International AI Safety Report. The document organizes AI safety research into a defence-in-depth framework with three areas: (1) Assessment — methods to measure AI system impacts, enhanced metrology, and third-party audit infrastructure; (2) Development — specification, training, and verification of trustworthy systems, including limiting AGI risk by constraining autonomy, generality, or intelligence; (3) Control — post-deployment monitoring, ecosystem oversight, societal resilience, and AI control setups robust to active subversion. Uniquely identifies areas of mutual interest where even competitors benefit from cooperation on shared risk thresholds, evaluation protocols, and safety benchmarks.