The International AI Safety Report lands with 30 nations behind it
Led by Yoshua Bengio, drawing on over 100 experts nominated by more than 30 countries. The broadest multilateral assessment of AI risk yet produced — and its central finding is a gap, not a hazard.
The 2026 International AI Safety Report has been published, led by Turing Award recipient Yoshua Bengio and drawing input from over 100 AI experts nominated by more than 30 nations. It is the broadest multilateral assessment of AI risk assembled to date.
Multilateral documents usually converge on the least objectionable statement available. This one converges somewhere more specific: alignment research has not kept pace with capability development, and the mechanisms for regaining that parity are not yet in place.
That is a claim about institutional capacity rather than about any particular model, and it is harder to dismiss for exactly that reason. A capability warning invites the response that the capability has not materialised. A capacity warning is checkable against what actually exists — how many people, how much compute, what evaluation infrastructure — and the answer does not depend on forecasting.
Its practical weight is as a citation. Thirty nations' nominated experts agreeing on a finding gives regulators a document to point at, which is what regulators need in order to act. Expect it in the footnotes of enforcement reasoning and in national AI strategies for the next two years.
It arrives in the same fortnight the EU demonstrated it will actually fine people. Assessment and enforcement moving together is unusual and probably not coincidental.
arXiv — Frontier AI Risk Management Framework in Practice: A Risk Analysis Technical Report v1.5 → · Cloud Security Alliance — The Alignment Gap: Control Failure Risk Before ASI →