DeepMind: Hassabis further shortens his AGI timeline
Demis Hassabis now estimates that AGI could arrive in two to three years, down from three to five years earlier in 2026. The Google DeepMind founder describes a near threshold while calling for better market regulation.
Demis Hassabis, founder of Google DeepMind and Nobel Prize winner in chemistry, now considers the arrival of AGI within two to three years possible. AGI, or artificial general intelligence, refers to a system capable of matching the full range of human cognitive abilities. The estimate tightens further, after an already lowered range earlier this year of three to five years. In this context, the head of AI at Google describes a shift that could affect research, products, and regulation much sooner than expected.
Une échéance qui se rapproche à vue d’œil
According to Matthias Bastian in The Decoder, Hassabis told the Wall Street Journal that "It's going to happen soon. My estimate keeps moving closer to the present." The phrase sums up the current mood in generative AI: timelines are sliding, but in the wrong direction for skeptics. At the start of the year, he was still talking about three to five years. Now, his window drops to two or three years, with a precise definition of AGI: a system that reaches the full cognitive range of the human brain.
To measure what he means by this, Hassabis relies on a test he calls the "Einstein test." The idea is simple in appearance: a model trained only on knowledge before 1901 should be able to reproduce Albert Einstein's 1905 work. If the answer becomes yes, he believes, a very different milestone from current uses, still focused on specialized tasks, will have been crossed. The nuance matters, because the debate is not only about raw power but about the real generalization of capabilities.
Un saut technologique, mais aussi un sujet de gouvernance
Hassabis compares the potential impact of AGI to "10 times the industrial revolution at 10 times the speed." The image is deliberately massive, but it clearly expresses the economic stakes: if such a technology arrives in the short term, it will not remain confined to lab demonstrations. It could affect knowledge professions, drug design, programming, and more broadly, how companies balance automation and human expertise.
The researcher has since refocused on Isomorphic Labs, Google's spin-off dedicated to drug discovery through AI. On the regulatory side, he advocates for an oversight council funded by industry and placed under government control, tasked with reviewing models before their release. He also warns that poor regulation can be worse than no regulation at all. In a sector where everyone wants to move fast, the message cuts short the usual choreography of promises: speed, yes, but not without brakes.
Ce que cette sortie dit du marché de l’IA
This new estimate is not proof that AGI will actually arrive in two years. However, it shows that some of the most advanced players in the sector are preparing for capabilities far broader than those of current assistants. For investors, labs, and regulators, the challenge becomes less about whether the race continues and more about deciding who sets the rules, and with what level of proof before deployment.
In practice, the issue goes beyond model performance alone. It touches on security, legal responsibility, intellectual property, and system control before deployment. At this stage, Hassabis does not speak of a firm date but of a window that is closing in. That is already precise enough to shift the landscape, without requiring champagne or hard hats.