Signal The Singapore Consensus on Global AI Safety Research Priorities (arXiv:2506.20702)

Summary

More than 100 AI experts from 11 countries agreed a shared set of AI safety research priorities, published as the Singapore Consensus following the Singapore Conference on AI. The document was authored by Yoshua Bengio together with 87 other contributors and builds on the International AI Safety Report, which Bengio also chaired and which is backed by 33 governments. The consensus emerged from the 2025 Singapore Conference on AI: International Scientific Exchange on AI Safety, held in April 2025. It organizes research priorities around a defence-in-depth model spanning three categories: development of trustworthy AI systems, assessment of risks, and control through monitoring and post-deployment intervention. The report states that building a trusted ecosystem is essential to helping people embrace AI with confidence while giving maximal space for innovation and avoiding backlash.