AGI Made Clear › Timelines
AGI timeline predictions, and how much they move
The useful thing about forecasts is not the dates. It is the track record of the people giving them, which is documented and rarely printed.
Every date below is real and sourced. So is every earlier date from the same person. Put those two facts next to each other and a forecast starts to look like what it is: a considered opinion that moves, not a measurement.
The clearest case: one man, one definition, five years in five months
Demis Hassabis (Google DeepMind) has said consistently that his definition of AGI has never changed, and there is no evidence it has. The main definitions in use disagree about what would settle the question. What changed, between January and June 2026, was the date attached to it.
March 2025“Five to ten years”Implies 2030-2035
July 2026“Probably only a few short years away”Implies about 2029-2031
One definition, five years of movement in five months →
That is a five-year revision in five months, from the person running one of the two most capable labs in the world, with the definition held fixed throughout. It is the strongest available argument for treating all of these numbers as opinion.
And the direction of travel is not uniform
Hassabis shortenedFrom “five to ten years” in January 2026 to “2030 plus or minus a year” in June 2026.
Amodei lengthenedFrom “as early as 2026” in October 2024 to 2028 in February 2026 - while describing his own exponential as having run “plus or minus a year or two” to plan.
Altman softened on impact, hardened on capabilityIn August 2026 he conceded he had been wrong about the speed of economic disruption: “the economy just has so much inertia.” In the same period OpenAI’s capability claims got more assertive, not less.
Suleyman withdrew from the questionHaving given “five to seven years” in December 2024, he now declines to give any figure and disputes that there is a race to win.
The claim we would make“Everyone is converging on the same date” is false, and so is “the timelines keep getting shorter”. Two of the four moved in each direction. What is actually happening is that individual forecasts are volatile, which is what you would expect of forecasts about something nobody has defined.
Predictions still circulating that are out of date
“Suleyman says AGI in five to seven years”That figure is from December 2024, on The Verge’s Decoder podcast. He has since stopped giving numbers at all and argues the race framing itself is wrong. Anything citing it as a current position is close to two years out of date.
“Amodei says AGI as early as 2026”From Machines of Loving Grace, October 2024. His current central estimate, given on the Dwarkesh podcast in February 2026, is 2028 - and he no longer uses the word AGI for it, preferring “powerful AI”.
“AGI is a few decades away, says Hassabis”A transcription error. An automatic transcript of his June 2026 Stanford appearance renders “2030” as “20, 30”, which at least one write-up read as two to three decades. Contemporaneous reporting confirms he said 2030.
