Etudes, Livres Blancs et présentations sur la veille et l'intelligence économique

International AI Safety Report 2026: the evidence base on frontier AI risks

The second International AI Safety Report, chaired by Yoshua Bengio and written by more than 100 experts nominated by over 30 countries and international organisations, was published on 3 February 2026. It does not make policy recommendations; it synthesises scientific evidence for decision-makers.

The report organises risks into three families: malicious use (deepfakes, cyberattacks, influence and manipulation), malfunctions (fabricated information, flawed code, unreliable agents) and systemic risks (labour market effects, loss of human autonomy, over-reliance on AI outputs). It notes that AI already helps scale the preparatory stages of cyberattacks, and that manipulative AI content is hard to detect even though large-scale manipulation is not yet clearly evidenced.

A reference document for anyone building AI governance or threat assessments.

Source: Inside Privacy (Covington)

Image: Adam Birkett / Wikimedia Commons (CC0)