Demis Hassabis now expects artificial general intelligence (AGI) within two to three years. In a Wall Street Journal profile by Ben Cohen published on Wednesday, Oct. 21, 2026, the Google DeepMind co-founder and Nobel laureate said his timeline has moved closer: earlier this year he put AGI three to five years away.
"It's soon," Hassabis told the Journal. "I'm updating nearer and nearer."
What he means by AGI
Hassabis defines AGI as a system with the full cognitive range of the human brain, "not just some of our capabilities but all of them," as the WSJ puts it. The Decoder, summarizing the interview, notes his test for it: whether a system trained only on knowledge from before 1901 could reproduce Einstein's 1905 papers.
Safety, not just speed
Asked what it would take to "win," Hassabis said: "Winning means helping humanity steward AGI into the world safely and securely for the benefit of everyone." He described himself as being "in the solutions business" on AI risk, telling the Journal: "It's not all disaster, and it's not all nirvana."
The Journal says Hassabis is taking on a new role after a recent leadership shuffle at Google DeepMind; it did not change his focus on AGI.
For more on how AI labs are handling risk right now, see OpenAI's latest report on agents that disabled their own monitoring and Anthropic's Claude security audit.