哈薩比斯支持發布前測試：FDA式AI監管是否不可避免
谷歌deepmind ceo demis hassabis支持ai模型的強制性發布前安全測試，模式類似於finra/fda協議。這標誌著行業自我監管的重大轉變。
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.