The Singapore Consensus on Global AI Safety Research Priorities
As AI capabilities rapidly evolve, there is vigorous global debate over how to keep AI safe, reliable and beneficial. To fully understand the extent of the risks and how to tackle these risks, there needs to be technical AI safety research areas that are broadly agreed to be valuable, and prioritised to enable more effective R&D efforts to drive safety and evaluation mechanisms to achieve a trusted ecosystem for AI use.
A key goal of the 2025 Singapore Conference on AI (SCAI): International Scientific Exchange on AI Safety is identifying and prioritising such research areas. Bringing together the global research community and building on the International AI Safety Report chaired by Yoshua Bengio, SCAI 2025’s outcome document – the Singapore Consensus on Global AI Safety Research Priorities – aims to demonstrate substantial consensus around identified important technical AI safety research domains. It will also serve as a foundation document to facilitate global, meaningful conversations amongst AI policymakers, scientists, the broader industry and public, improve understanding of risk management, and spur international research collaboration.
The Singapore Consensus is a living document and continues to welcome views from the global research community.
Explore the interactive version of the report
Click on the link below to take a closer look at the report in an interactive format.