DeepMind CEO says AGI may arrive within a few years

Google DeepMind CEO Demis Hassabis said artificial general intelligence could be only a few years away, and he is now pushing for a new U.S. oversight system before the most powerful AI models reach the public. For investors watching the AI boom, the point is straightforward: the people building frontier models are also asking for rules that could shape how quickly those systems are shipped.

In a post published Tuesday on X, Hassabis said AGI may arrive before the end of the decade. AGI, short for artificial general intelligence, refers to software that can learn, reason, and perform a wide range of tasks at or above human level.

Hassabis argued that AGI should be viewed as a civilizational shift rather than another consumer-tech upgrade. He wrote that its importance could be closer to the discovery of electricity or fire than to the internet or mobile computing.

He also described the current moment as the beginning of a new era for humanity, saying future generations may see this period as the early stage of what he called the singularity, a term often used for a point when technological change accelerates beyond ordinary human control.

Why Hassabis wants pre-release AI testing

Hassabis paired the big prediction with a regulatory proposal. He called for the creation of a U.S. Frontier AI Standards Body that would test advanced AI systems before companies deploy them.

His proposed body would be modeled on the Financial Industry Regulatory Authority, or FINRA, the private organization that oversees U.S. brokerage firms under federal supervision. Hassabis said the AI version should operate as a federally supervised public-private partnership, funded mainly by the AI industry and staffed by independent technical experts and representatives from the open-source community.

The job would be to evaluate frontier AI models, meaning the most capable systems being developed, before release. Hassabis said that kind of testing could later become mandatory for the strongest models.

The mechanism matters. Pre-release testing would put an outside review step between model development and public deployment. If adopted, that could affect product timelines, compliance costs, and the way AI companies disclose risks, though Hassabis did not provide estimates for any financial impact.

Risks are moving faster than institutions

Hassabis said today’s frontier systems already create cybersecurity concerns. He warned that future models could raise additional biological, nuclear, and national security risks.

He also pointed to the challenge of keeping control as AI systems become more agentic. In AI, agentic systems are programs that can pursue goals, make decisions, and take actions with less direct human instruction. Hassabis said stronger technical safeguards will be needed as systems become more capable and potentially able to improve themselves.

His comments add to a broader push from AI executives for stronger oversight of advanced systems. In May 2023, OpenAI CEO Sam Altman told the U.S. Senate Committee on the Judiciary that the federal government should create an agency to license powerful AI systems and require independent safety audits.

Anthropic CEO Dario Amodei has also warned that human-level AI may arrive sooner than governments expect. In January 2026, Amodei said such systems could emerge within one to five years, according to Decrypt.

Hassabis made a similar timing call earlier this year. In June, he said AGI could arrive by 2030 and warned that society did not have much time to prepare, according to Decrypt.

The latest proposal puts a sharper policy frame around that timeline. Hassabis is arguing that if AGI is close, testing rules need to be built before the technology reaches its most powerful form.

This story draws on original reporting from Decrypt.