Hassabis: AGI could rival electricity, urges U.S. AI body

DeepMind CEO Demis Hassabis wrote in a blog post on X that artificial general intelligence is “probably only a few short years away” and compared its potential impact to the discovery of electricity or fire. The post was published Tuesday and framed the technology as capable of reshaping human civilisation.

Hassabis warned that AI capabilities are advancing faster than society’s ability to understand and manage the risks. He pointed to existing cybersecurity threats linked to today’s most capable models and said future systems could introduce biological, nuclear and other national security risks. He added that as models become more agentic and able to improve themselves, stronger technical safeguards will be needed to keep humans in control.

To address those concerns, Hassabis proposed creating a U.S. Frontier AI Standards Body modeled on the Financial Industry Regulatory Authority, a privately run regulator for brokerage firms. The plan envisions a federally supervised public-private partnership funded mainly by the AI industry and staffed with independent technical experts and open-source representatives.

The proposed body would conduct dynamic, adaptable pre-release testing of frontier AI models. The proposal allows for testing to become mandatory for the most capable systems, with the goal of identifying risks before models are widely deployed.

Hassabis argued the United States is well placed to lead the development of such a framework given its economic and technical standing, and he urged action during what he described as a limited window to agree on common standards before AGI arrives. He wrote that the future is “not yet written” and called for steps to shape the technology for broad benefit.

The proposal follows recent industry and government activity on advanced AI oversight. In January 2026, Anthropic’s chief executive forecast that human-level AI could appear within one to five years. In 2023, the chief executive of another major AI developer urged the creation of a federal agency to license powerful AI systems and require independent safety audits. Last month, the White House issued an executive order setting a voluntary framework for reviewing advanced AI models prior to release.

AGI refers to systems that can understand, learn and perform a wide range of tasks at or above human level, rather than narrow systems trained for specific tasks. Frontier models are the most capable current systems; supporters of pre-release review say such testing can reveal risks before public deployment, while critics raise questions about governance, enforcement and the effect of mandatory checks on the pace of innovation.