Google DeepMind CEO Demis Hassabis has issued a warning that artificial general intelligence (AGI) could be achieved within the next three to four years. He defines AGI as systems capable of matching or exceeding human performance across nearly all cognitive tasks, emphasizing that the current era is a defining period for humanity.

In response to this rapid advancement, Hassabis is calling for the establishment of a global, independent oversight body modeled after FINRA, an independent regulatory organization for the financial industry. He suggests this body should be primarily funded by the AI industry and staffed by technical experts to set mandatory safety benchmarks for frontier AI models. While initially voluntary, Hassabis proposes that safety testing should eventually become a mandatory requirement for all advanced AI developers.

Hassabis argues that current competitive pressures between corporations and nations are distracting from critical safety considerations. He noted that even leading researchers currently struggle to fully interpret the internal operations of the most advanced models, making their future behaviors difficult to predict. The proposal specifically addresses potential risks in high-stakes fields such as cybersecurity, biology, and nuclear research, where increasingly autonomous systems could pose significant threats if not properly contained.

Despite his emphasis on these hazards, Hassabis maintains an optimistic long-term view of AGI. He compared its potential societal impact to foundational discoveries like electricity or fire rather than the internet. He envisions that if developed responsibly, AGI could significantly accelerate advancements in fields like drug discovery, clean energy, and materials science. He cautioned, however, that the decisions made in these coming years will determine whether this technology results in widespread prosperity or leads to avoidable risks.

Google DeepMind CEO Demis Hassabis has issued a warning that artificial general intelligence (AGI) could be achieved within the next three to four years. He defines AGI as systems capable of matching or exceeding human performance across nearly all cognitive tasks, emphasizing that the current era is a defining period for humanity.

In response to this rapid advancement, Hassabis is calling for the establishment of a global, independent oversight body modeled after FINRA, an independent regulatory organization for the financial industry. He suggests this body should be primarily funded by the AI industry and staffed by technical experts to set mandatory safety benchmarks for frontier AI models. While initially voluntary, Hassabis proposes that safety testing should eventually become a mandatory requirement for all advanced AI developers.

Hassabis argues that current competitive pressures between corporations and nations are distracting from critical safety considerations. He noted that even leading researchers currently struggle to fully interpret the internal operations of the most advanced models, making their future behaviors difficult to predict. The proposal specifically addresses potential risks in high-stakes fields such as cybersecurity, biology, and nuclear research, where increasingly autonomous systems could pose significant threats if not properly contained.

Despite his emphasis on these hazards, Hassabis maintains an optimistic long-term view of AGI. He compared its potential societal impact to foundational discoveries like electricity or fire rather than the internet. He envisions that if developed responsibly, AGI could significantly accelerate advancements in fields like drug discovery, clean energy, and materials science. He cautioned, however, that the decisions made in these coming years will determine whether this technology results in widespread prosperity or leads to avoidable risks.