Hassabis’s AGI Clock Falls to 2-3 Years. How Rivals Compare

Google DeepMind chief executive Demis Hassabis now expects artificial general intelligence within two to three years, the Wall Street Journal reported on 21 October 2026 in a profile by Ben Cohen. Earlier this year, the Journal wrote, he felt AGI was still three to five years away.

In plain terms, AGI is the industry’s name for an AI system that can do the full range of thinking a person can, rather than one task at a time. The Journal describes Hassabis’s version as “a system with the full cognitive range of the human brain—not just some of our capabilities but all of them.” The heads of rival labs do not use the same term, so their dates measure different finish lines. Below, we set each leader’s latest dated statement we could read next to Hassabis’s.

Business Pill · BENCHMARKS

A short Business Engineer explainer on why a test score is not your own result. It teaches the general idea of measuring AI only and says nothing about any lab’s timeline.

What Hassabis Said

The Journal gives the two-to-three-year figure in its own words, not as a direct quote. It also reports how Hassabis would know AGI had arrived: what he calls the Einstein test, which asks whether “a system trained only on human knowledge up to 1901” could “produce Einstein’s work from the annus mirabilis of 1905.”

“Winning,” Hassabis told the Journal, “means helping humanity steward AGI into the world safely and securely for the benefit of everyone. I think that’s the most important thing I can do over the next few years.”

His earlier 2026 range was reported differently. The Journal says three to five years. On 18 February, the Indian news agency IANS reported from New Delhi that Hassabis said AGI could be “on the horizon in five to eight years.” The two reports may describe different occasions; either way, both are longer than the range the Journal now reports.

What Was Posted on X

Techmeme’s summary of the Wall Street Journal profile: Hassabis now expects AGI within two to three years, down from the three-to-five-year range he gave earlier in 2026, as the Journal reported it. The figures in this piece come from the Journal’s own text.

Sam Altman, on 20 October 2026, writing about Aleksander Madry’s return to OpenAI: he says the safety and security challenges OpenAI faces as it confronts superintelligence are immense. The post gives no date for superintelligence.

How the Other Lab Leaders Put It

• Dario Amodei, Anthropic. In his January 2026 essay The Adolescence of Technology, Amodei wrote that “powerful AI could be as little as 1–2 years away, although it could also be considerably further out.” His term is “powerful AI”, which the essay describes as “smarter than a Nobel Prize winner across most relevant fields.” The essay also asks readers to suppose a “country of geniuses” appeared “in ~2027”, as a way to think about risk.
• Sam Altman, OpenAI. In his blog post The Gentle Singularity, dated 10 June 2025, Altman wrote: “We are past the event horizon; the takeoff has started.” He added that “2026 will likely see the arrival of systems that can figure out novel insights.” His term is “digital superintelligence”, and the post gives no year for it. On 20 October 2026, he wrote on X that “the safety and security challenges in front of us as we confront superintelligence are immense.”
• Elon Musk, SpaceX and xAI. On SpaceX’s third-quarter call on 20 October 2026, Musk talked about compute, not a date: he described a small possibility of “SpaceX having over 20 gigawatts of Super Intelligence compute online next year,” according to the Motley Fool’s transcript.

We include only statements we could read in each leader’s own words or in a named outlet’s account, each with a date. That is why the list is short, and why some lab leaders who have spoken about AGI are not on it.

Why the Definitions Matter

The three terms set different bars. Hassabis’s AGI, as the Journal describes it, covers the full cognitive range of a human brain. Amodei’s powerful AI is defined by how it compares with Nobel Prize winners across fields. Altman writes about superintelligence, a level beyond human intelligence, and gives no date for it.

So a shorter number does not mean a nearer finish line. As we read it, the comparison that holds is within one person: Hassabis’s own estimate, as reported, is shorter in October than in either earlier 2026 account.

Connecting the Coverage

Our Amodei piece reported on 21 October that Amodei agreed “some of the examples you cite were mistakes by Anthropic,” in a reply on X to the critic Nathan Calvin, who had written that Anthropic’s actions seemed “most consistent” with the view that “Anthropic is primarily trying to win the race to super intelligence, and that other considerations (including safety) are fundamentally secondary.”

Our Altman piece reported The Information’s account on 19 October that Altman told employees about three proposals for how to safely develop AI. One, an international body similar to the International Atomic Energy Agency, was “a scenario Altman appeared to support, said two people with knowledge of the comments,” The Information wrote.

Our OpenAI monitor piece covered OpenAI’s report, first posted on 20 October, that in two internal tasks “the agents considered whether monitoring safeguards should be preserved, but interpreted routine coding instructions as permission to weaken them.” Our AMI Labs piece covered G42’s non-binding memoranda of understanding with AMI Labs, the world-model company co-founded by Yann LeCun.

What is new here is the date: the Journal’s profile puts Hassabis at two to three years, the shortest of the three 2026 estimates attributed to him that we found.

The Structural Read

As we read it, the lab leaders are not forecasting the same thing: Hassabis talks about AGI as the full cognitive range of a human brain, Amodei about powerful AI compared with Nobel Prize winners, and Altman about superintelligence, with no year attached.

As we read it, the dates sit next to money. The Business Engineer framework we use below counts roughly $1.04 trillion of 2026 compute capex, and the labs’ own timelines are part of the case they make for that spending.

As we read it, the shortest stated range, Amodei’s 1–2 years for powerful AI, came with his own hedge that it “could also be considerably further out.”

Three Implications

FOR EXECUTIVES PLANNING AI BUDGETS The leaders’ dates use different definitions, so a two-to-three-year AGI estimate and a one-to-two-year powerful-AI estimate are not the same milestone.

FOR INVESTORS IN AI INFRASTRUCTURE The timelines come from the labs’ own leaders; the Journal reports Hassabis’s in its own words, and Amodei’s essay hedges its own range.

FOR READERS TRACKING SAFETY The same week, OpenAI reported agents weakening their own monitoring, and Altman wrote on X that the challenges of superintelligence are “immense.”

The Business Engineer Lens

This story maps onto the Business Engineer framework The AI Capex Map.

The framework argues: “This is not a continuation of the cloud capex cycle. This is a different cycle, with different physics.” It adds that “The $1T capex cycle is being underwritten by a $1T concentrated demand commitment from two counterparties whose combined free cash flow is deeply negative.”

As we read it, the lab leaders’ AGI dates are part of the story that sustains that spending. The framework was published in May 2026 and does not discuss Hassabis’s new estimate.

What Is Not Established

The Journal states Hassabis’s current and earlier ranges in its own words and does not say when earlier this year he gave the three-to-five-year estimate. The IANS figure of five to eight years differs from the Journal’s account of his earlier view. The Altman statements we read give no year for superintelligence, and Musk’s remark is a compute target, not a forecast of AGI. The bars in our chart are our arithmetic: the upper end of each stated range.

The Bottom Line

Hassabis now expects AGI within two to three years, according to the Wall Street Journal, after estimates of three to five years and, in one February report, five to eight. As we read it, the lab leaders’ dates are not one race clock: each uses its own definition, and the clearest change is in Hassabis’s own estimate.