DeepMind CEO Calls for U.S. AI Safety Review Framework as AGI 'Only Years Away'

According to DeepMind CEO Demis Hassabis, on Tuesday, July 14, he published a lengthy statement on X warning that artificial general intelligence (AGI) is only years away and calling for the U.S. to establish a frontier AI safety review mechanism. Hassabis, who co-founded DeepMind and developed AlphaGo and AlphaFold, proposed a new standards body modeled after FINRA (Financial Industry Regulatory Authority) to conduct pre-release safety evaluations of advanced models within 30 days of submission.

The proposal emphasizes risks including cybersecurity threats and potential biological hazards as AI capabilities advance. Hassabis noted that safety warning signals are now appearing almost monthly, citing examples including OpenAI's GPT-5.6 Sol model, which faced U.S. government review for its cybersecurity capabilities, and Anthropic's models that underwent export control restrictions. He stressed that as AI systems gain recursive self-improvement abilities, robust safety guardrails become critical to maintain control.