Geoffrey Hinton Proposes FDA-Style Safety Approval Mechanism for AI Models

Turing Award winner Geoffrey Hinton proposed that AI companies should be legally required to prove model safety to regulators before public deployment, drawing a direct comparison to FDA drug approvals. He warned that rapid advancements driven by recursive self-improvement could significantly escalate AI safety risks within just one to two years. Hinton's call amplifies the ongoing global debate over mandatory government oversight and pre-deployment safety testing for frontier AI models. Requiring pharmaceutical-grade evaluation standards could fundamentally reshape development timelines and force companies to invest heavily in safety verification. Hinton noted that pharmaceutical drug approvals cost around a billion dollars in safety testing, suggesting AI safety evaluation demands comparable investment and oversight. He specifically cited recursive self-improvement—where AI tools are used to optimize and develop newer AI systems—as a primary accelerator of exponential risk.