Second Key Update of the International AI Safety Report Released

On November 25, 2025, the second Key Update of the International AI Safety Report was published in Montreal. This report provides a current perspective on managing risks and implementing technical mitigations for general-purpose AI. Chaired by Turing Award winner Yoshua Bengio, it assembles insights from over 100 international experts and receives backing from more than 30 countries and organizations, including the EU, OECD, and UN.

The second update highlights advances in training models to resist attacks, though gaps remain, as sophisticated hackers exploit vulnerabilities in about 50% of cases. It also notes the narrowing gap in open-source developments compared to industry leaders, which democratizes access to AI models. Furthermore, while industry commitment to safety has doubled, the actual effectiveness of these measures is still unclear.

This update ensures policymakers have current data to guide evidence-based decision-making ahead of the anticipated publication of the complete International AI Safety Report in early 2026.