Votre vie privée
Français

Organismes ayant un pouvoir

International AI Safety Report (Bengio-chaired)

Type
Multilateral body
Lieu
International — International (UK AI Security Institute hosts the Secretariat)
Dernière vérification
2026-09-22
Prochaine vérification
2027-03-21

Pas encore traduit — affiché en anglais.

Comment les joindre

  • General enquiries · Email address
    Vérifié · 2026-09-22

    The only published contact address

    Verified by direct fetch of the About page.

  • Secretariat-coordinated review process · Public consultation
    Vérification expirée · 2026-09-22

    Ce canal est conservé uniquement comme référence. Sa vérification est absente, date de plus de 180 jours, ou la période de contact n'est pas ouverte actuellement ; ne vous y fiez pas avant une nouvelle vérification.

    Draft review by civil society and industry

    The About page states the UK Secretariat 'coordinates a review process that gathers feedback from civil society groups and industry leaders'. There is NO published open call for evidence and no public submission form — participation is by invitation into the review round.

  • Homepage · Homepage
    Non contrôlé

Ce qu'il fait

Mandated by the Bletchley summit; chaired by Yoshua Bengio. 2026 edition published 3 Feb 2026 (arXiv 2602.21012) ahead of the India AI Impact Summit, with an Executive Summary and an Extended Summary for Policymakers. Backed by an Expert Advisory Panel of the 30 Bletchley countries plus the EU, OECD and UN. Synthesises evidence on general-purpose AI capabilities, risks and technical mitigations; it does not make policy recommendations.

Évaluation franche

Prestigious but not open. There is one inbox and an invitation-only review round. A well-evidenced technical finding emailed to enquiries@ has a plausible path to an author; a general policy opinion does not. The higher-yield route is to get the underlying work cited in the literature the authors survey, since the Report is a synthesis of published evidence.

Préoccupations couvertes

Comprehensive by design: malicious use (cyber, bio, manipulation), malfunctions/reliability, systemic risks (labour, concentration, environment), and loss of control are all treated. Notably more balanced across near-term and catastrophic risk than its critics assume.

Canal gouvernemental

Very strong — it is the designated evidence base for the AI summit series and is formally received by ~30 governments plus the EU, OECD and UN.

Sources

Une erreur ?