The International AI Safety Report anchors a 30-country consensus as enforcement begins
The second International AI Safety Report — led by Yoshua Bengio, authored by over 100 experts and backed by more than 30 countries and international organisations — has become the shared evidence base regulators reach for as the EU, China and US regimes diverge on everything else.
Governments that agree on almost nothing about AI regulation have converged on a single technical baseline. That is unusual and useful: when the EU enforces transparency, China fines companion-AI providers and Washington deadlocks, a common description of what the risks actually are keeps the arguments about remedies from becoming arguments about reality.
The scale is what gives it force. A hundred-plus authors and thirty-plus backing states make it hard to dismiss as any one bloc's position, and its findings — particularly on evaluation awareness and the weak link between pre-deployment testing and deployed behaviour — have been cited across otherwise incompatible regimes.
Its limitation is equally clear. The report describes; it does not bind. Its influence depends on regulators choosing to use it, and the gap between shared diagnosis and divergent enforcement is precisely where multinational compliance teams now live.
International AI Safety Report — International AI Safety Report 2026 → · Wikipedia — International AI Safety Report →