Hassabis tells the Wall Street Journal he now puts AGI at two to three years

Demis Hassabis now expects artificial general intelligence within two to three years, closer than the three-to-five-year range he was using earlier in 2026. He said so in a Wall Street Journal Magazine profile by Ben Cohen, published October 21 and carried the same day by Mint. The Verge reported the same shift that afternoon, pointing at the Journal profile.

This is his forecast, not a delivery date. Nothing in the profile is a Google commitment to ship a system on that clock. The Wall Street Journal page itself did not load for this desk; the quotations below are from Mint’s copy of Cohen’s article, which carries the Journal byline. The Verge quotes the same two lines about the timeline and about “winning.”

What he means by the term

Hassabis describes AGI, in Cohen’s paraphrase, as a system with the full cognitive range of a human brain, not a subset of human skills. Asked about the shorter clock, he told Cohen: “It’s soon. I’m updating nearer and nearer.”

He also described a test he uses for the idea. Cohen writes that Hassabis will treat a system as having reached that bar if, trained only on human knowledge up to 1901, it could produce Einstein’s work from 1905. Cohen adds that Hassabis says the field is still a few years from a system that can invent modern physics, and that this is sooner than he would have predicted when he won the Nobel Prize.

On what “winning” the race would mean, Hassabis told Cohen: “Winning means helping humanity steward AGI into the world safely and securely for the benefit of everyone.” The Verge reports that sentence as well.

The profile says that after running Google DeepMind, Hassabis recently became chief scientist of Alphabet, Google’s parent, and chair of DeepMind, and that he is also chief executive of Isomorphic Labs, the drug-discovery company he started inside Google. Those titles are as the profile states them. This desk did not check a separate Alphabet filing.

A forecast, and a warning about rules

Hassabis is quoted arguing against writing detailed rules for systems that are not yet understood. “What would be wrong is to try and regulate things now that we don’t even fully understand,” he told Cohen. “I think you’re almost certainly going to get the wrong regulation — and wrong regulation is worse than no regulation.” That is his view. The profile also says that over the summer he wrote a proposal for an industry-funded body, under federal oversight, that would test frontier models before release. The profile does not include the proposal’s text.

Cohen writes that Hassabis wants to use a future AGI system, together with human scientists, to work on “the next level of physics.” He quotes Hassabis on medicine as well: asked whether AI-designed drugs face an insurmountable obstacle, Hassabis said, “I don’t see any insurmountable obstacles currently. I see hard things, but not impossible things.”

Readers should treat every year-count in the profile as one person’s judgment, offered in an interview. The piece does not present a benchmark result, a model card, or a date on a roadmap.

This report is not legal, financial, or health advice. See the disclaimer.