Sir Demis Hassabis advocates for a mandatory preflight testing system for large-scale AI models to ensure market readiness and reliability.

Sir Demis Hassabis, the CEO of Google DeepMind, has publicly endorsed a preflight testing system for artificial intelligence models before they are deployed in the US market. The proposed system would require frontier labs to share models with an independent standards body for review up to 30 days prior to release. Once the assessment protocol is proven effective, the administration could formalize the process, making it mandatory for models to pass the test before reaching consumers. Hassabis emphasized the importance of involving independent technical experts in the process to avoid potential conflicts of interest, especially as the government considers taking equity stakes in AI companies. This move follows a period of widespread hostility toward AI regulation in Washington. While the administration has already begun implementing some form of preflight testing, Hassabis argues that a formal, transparent, and independent model—similar to the FINRA model—will improve the global AI landscape. The proposal aims to identify critical vulnerabilities early, ensuring that the most robust models reach the public.

Sources