Hassabis Now Puts AGI Two to Three Years Away, Down From Three to Five Earlier This Year

Demis Hassabis, Google DeepMind founder and Nobel laureate in chemistry, now says AGI could arrive within two to three years, according to The Decoder's report on his Wall Street Journal interview. Earlier this year, he put the estimate at three to five years.

"It's going to happen soon. My estimate keeps moving closer to the present," Hassabis told the WSJ.

How Hassabis defines AGI

Hassabis defines AGI as a system that matches the full cognitive range of the human brain. That is a stricter bar than definitions built around economic output or performance on a set of tasks.

His practical litmus test is the "Einstein test": could a system trained only on knowledge available before 1901 independently reproduce Einstein's 1905 papers? No benchmark scores, evaluation protocol, or model has been published for this test. It is a thought experiment, not a measurable metric.

Impact claims

Hassabis has compared AGI's expected impact to "10 times the industrial revolution at 10 times the speed." This is his own characterization. No supporting economic modeling or data was cited in the source material.

Safety and regulation

On safety, Hassabis recently proposed an industry-funded oversight board operating under government control. The board would vet models before release. He also warned that bad regulation is worse than none.

The proposal's details are not disclosed in the source: funding mechanism, legal authority, which models would be covered, and what a pass/fail standard would look like.

Role at Google DeepMind

The Decoder reports that Hassabis has stepped down from Google DeepMind and is now focused on Isomorphic Labs, Google's drug discovery AI spin-off. We could not independently confirm this from the source material, and the report describes him as the DeepMind founder rather than as its current leader. Readers should treat his current role as unverified until Google or Isomorphic confirms it.

Not a model release

This is a forecast, not a product announcement. No new model, context window, pricing, benchmark results, or training cutoff were disclosed.

What this means

The notable change is the direction and speed of the revision. Hassabis has moved his estimate by roughly one to two years in a matter of months. He is one of the most prominent researchers to use a strict, brain-level definition of AGI, so the shift is a meaningful signal about how frontier lab leadership reads internal progress.

It is still a forecast. The Einstein test has no published methodology, so there is no way to verify progress toward it from outside. The two-to-three-year window cannot be falsified until it expires, and the definition is demanding enough that any claimed arrival will invite disputes over whether it counts.

The oversight-board idea may matter more in the near term. A proposal for pre-release vetting, funded by industry but controlled by government, sits between voluntary lab commitments and direct statutory regulation. Without details on scope and enforcement, it is a position statement rather than a workable framework. Combined with a shortened timeline, it suggests that senior researchers are pushing for governance structures to be in place before the capabilities they predict.