In January 2025, the “International AI Safety Report,” the world’s first comprehensive report integrating existing literature and exploring the risks and capabilities of advanced AI systems, was officially released. This landmark report was led by Turing Award-winning computer scientist Yoshua Bengio, bringing together contributions from 96 AI experts across 30 countries, with the aim to establish a solid foundation for international consensus on advanced AI risks.
The “International AI Safety Report” was formally presented at the Paris AI Action Summit in February 2025, with an interim version previously released at the Seoul AI Summit in May 2024. The report originated from the consensus of the 2023 Bletchley AI Safety Summit, jointly initiated by 30 countries and supported by expert advisory panels from the United Nations, European Union, and Organisation for Economic Co-operation and Development (OECD), truly realizing the summit’s vision of internationality, independence, and inclusivity.
Kwan Yee Ng, Head of International AI Governance at Concordia AI, served as a member of the writing group, contributing to the report’s drafting and refinement. Concordia AI provided comments and feedback on the overall content of the report and offered editorial and revision suggestions for the Chinese translation.
As the report chair, Bengio noted that since the Bletchley Summit, “the capabilities of general-purpose AI…have increased further,” including “[new models] have shown markedly better performance at tests of programming and scientific reasoning” while “many companies are now investing in the development of general-purpose AI ‘agents’ – systems which can autonomously plan and act to achieve goals with little or no human oversight.” This complete report not only “further improve(d) the scientific rigor of all sections” but also “added discussion of additional topics such as open-weight models” and “restructured the report to be more relevant to policymakers.”
Although the Report consolidates the scientific community’s current scientific understanding and consensus on general-purpose AI, the Report still emphasizes that “as the expert authors of this report, we still have many disagreements, large and small, regarding the capabilities, risks, and risk mitigation measures of general-purpose AI. However, we believe this report is crucial for enhancing our collective understanding of this technology and its potential risks. We hope this report will help the international community reach broader consensus on general-purpose AI and more effectively mitigate its risks, enabling people to safely enjoy its many potential benefits. We look forward to continuing this effort.”