AI researcher and critic Gary Marcus wrote on July 14, 2026 that Google DeepMind CEO Demis Hassabis has publicly endorsed preflight safety testing for AI models before deployment, an approach Marcus has long advocated modeled on the FDA’s pharmaceutical review process. Hassabis proposed that frontier labs voluntarily submit models for independent expert review up to 30 days before release, with mandatory compliance potentially following once the assessment protocol is proven effective. Marcus argues the proposal’s value depends on involving independent technical experts rather than leaving the process solely to government and corporate decision-makers.
