Hassabis氏、事前安全テストを支持――FDA型のAI規制は不可避か

demis hassabis publicly proposed a framework where frontier labs voluntarily submit models to an independent standards body for safety testing up to 30 days before release. he advocated for a model resembling finra or fda protocols, where successful assessments would eventually become mandatory for us market deployment. this move counters recent washington hostility towards regulation and addresses conflict of interest concerns by involving independent technical experts alongside government and big tech. if implemented, this system would standardize risk mitigation and fundamentally alter how ai models reach the public.