The Singapore Consensus on Global AI Safety Research Priorities
@misc{bengio2025singaporeconsensusglobalai,
title={The Singapore Consensus on Global AI Safety Research Priorities},
author={Yoshua Bengio and Tegan Maharaj and Luke Ong and Stuart Russell and Dawn Song and Max Tegmark and Lan Xue and Ya-Qin Zhang and Stephen Casper and Wan Sie Lee and Sören Mindermann and Vanessa Wilfred and Vidhisha Balachandran and Fazl Barez and Michael Belinsky and Imane Bello and Malo Bourgon and Mark Brakel and Siméon Campos and Duncan Cass-Beggs and Jiahao Chen and Rumman Chowdhury and Kuan Chua Seah and Jeff Clune and Juntao Dai and Agnes Delaborde and Nouha Dziri and Francisco Eiras and Joshua Engels and Jinyu Fan and Adam Gleave and Noah Goodman and Fynn Heide and Johannes Heidecke and Dan Hendrycks and Cyrus Hodes and Bryan Low Kian Hsiang and Minlie Huang and Sami Jawhar and Wang Jingyu and Adam Tauman Kalai and Meindert Kamphuis and Mohan Kankanhalli and Subhash Kantamneni and Mathias Bonde Kirk and Thomas Kwa and Jeffrey Ladish and Kwok-Yan Lam and Wan Lee Sie and Taewhi Lee and Xiaojian Li and Jiajun Liu and Chaochao Lu and Yifan Mai and Richard Mallah and Julian Michael and Nick Moës and Simon Möller and Kihyuk Nam and Kwan Yee Ng and Mark Nitzberg and Besmira Nushi and Seán O hÉigeartaigh and Alejandro Ortega and Pierre Peigné and James Petrie and Benjamin Prud'Homme and Reihaneh Rabbany and Nayat Sanchez-Pi and Sarah Schwettmann and Buck Shlegeris and Saad Siddiqui and Aradhana Sinha and Martín Soto and Cheston Tan and Dong Ting and William Tjhi and Robert Trager and Brian Tse and Anthony Tung K. H. and Vanessa Wilfred and John Willes and Denise Wong and Wei Xu and Rongwu Xu and Yi Zeng and HongJiang Zhang and Djordje Žikelić},
year={2025},
eprint={2506.20702},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2506.20702},
}
June 24, 2025
Abstract
Rapidly improving AI capabilities and autonomy hold significant promise of transformation, but are also driving vigorous debate on how to ensure that AI is safe, i.e., trustworthy, reliable, and secure. Building a trusted ecosystem is therefore essential -- it helps people embrace AI with confidence and gives maximal space for innovation while avoiding backlash. The "2025 Singapore Conference on AI (SCAI): International Scientific Exchange on AI Safety" aimed to support research in this space by bringing together AI scientists across geographies to identify and synthesise research priorities in AI safety. This resulting report builds on the International AI Safety Report chaired by Yoshua Bengio and backed by 33 governments. By adopting a defence-in-depth model, this report organises AI safety research domains into three types: challenges with creating trustworthy AI systems (Development), challenges with evaluating their risks (Assessment), and challenges with monitoring and intervening after deployment (Control).
