Research
Demis Hassabis Moved His AGI Forecast to 2029. That Is the Most Consequential Prediction in Science Right Now.
At Google I/O 2026, DeepMind CEO Demis Hassabis tightened his AGI timeline to 2029–2030, calling the current agentic era 'a practice run' and describing humanity as standing 'in the foothills of the singularity.' The forecast carries weight that no other public AGI prediction does.
By Elvin C ·
Demis Hassabis has won a Nobel Prize in Chemistry for his work on protein structure prediction. He leads Google DeepMind, the organisation that produced AlphaFold, AlphaGo, and Gemini. When he makes a public prediction about the timeline for artificial general intelligence, it is not a marketing claim or a speculative forecast from an outsider. It is the considered view of the person who has arguably done more than anyone alive to advance the state of AI capability — and who has more visibility into the current trajectory of frontier research than almost anyone outside the major labs.
At Google I/O on 20 May 2026, Hassabis said: 'When we look back at this time, I think we all realise that we were standing in the foothills of the singularity. It will be a profound moment for humanity.' In a subsequent interview with Axios, he tightened his AGI timeline to 2029 or 2030 — a significant revision from his previous estimate of 'shortly after 2030,' and from his estimate a year earlier of 2030 to 2035. The acceleration of the timeline reflects what Hassabis described as the rapid maturation of AI agents: 'We can see agents really happening now and imagine what they will be in another year, and how useful they'll be.'
'When we look back at this time, I think we all realise that we were standing in the foothills of the singularity. It will be a profound moment for humanity.' — Demis Hassabis, Google I/O 2026
Why This Forecast Is Different
AGI timeline predictions are common in the AI industry, and most of them deserve to be treated with scepticism. They are often made by people with financial incentives to generate excitement, by researchers whose domain expertise is narrow, or by commentators who are extrapolating from publicly available benchmarks without access to the internal research that drives frontier capability. Hassabis's forecast is different on all three dimensions. He has no obvious incentive to accelerate the public timeline — if anything, a shorter AGI timeline creates regulatory and reputational pressure on DeepMind. His domain expertise is unusually broad, spanning reinforcement learning, protein structure prediction, and large language models. And his access to frontier research at DeepMind is among the best in the world.
The forecast also comes from someone who has historically been conservative about AGI timelines. In 2022, Hassabis was describing AGI as 'decades away.' In 2024, he moved to 'within a decade.' Now, in 2026, he is saying 2029 is possible. Each revision has been in the direction of acceleration, and each has come after observing capability jumps that were not anticipated by the previous forecast. The pattern suggests that the revisions are driven by empirical observation rather than by a desire to generate headlines.
The Agentic Era as a Practice Run
Hassabis described the current agentic era — the period of AI agents that can take multi-step actions, use tools, and operate autonomously across extended tasks — as 'a bit like a practice run' for AGI. This framing is significant. It implies that the capabilities being developed and deployed today are not the destination but a stepping stone, and that the transition from current agents to AGI is a matter of degree rather than kind. The practice run framing also suggests that the infrastructure, safety frameworks, and governance mechanisms being built for current AI agents will need to be substantially upgraded before AGI arrives.
The comparison to other AGI forecasts is instructive. Ilya Sutskever, the former OpenAI cofounder who now leads Safe Superintelligence Inc., made a wide estimate of 2030 to 2045 in late 2025. Sam Altman has described AGI as potentially arriving 'within the next few years' in various public statements. Dario Amodei of Anthropic has been more cautious, focusing on 'powerful AI' rather than AGI as the near-term milestone. Hassabis's 2029 forecast is the most specific and the most credible, given his track record and his access to frontier research.
What AGI in 2029 Would Actually Mean
The definition of AGI remains contested. The most common working definition — a system that can perform any intellectual task that a human can perform — is broad enough to encompass a wide range of capabilities, and different researchers draw the line in different places. Hassabis has generally defined AGI in terms of general problem-solving ability across domains, rather than in terms of specific benchmark performance. By this definition, AGI would be a system that could, in principle, make scientific discoveries, write software, conduct research, and reason about novel problems without domain-specific training.
If AGI arrives in 2029 or 2030, the implications for scientific research are particularly significant. Hassabis has described drug discovery and materials science as the domains where AGI would have the most immediate impact. AlphaFold's impact on structural biology — which contributed to Hassabis's Nobel Prize — provides a template: a single AI system that solved a problem that had resisted decades of human effort, and that immediately accelerated research across the entire field. An AGI system would, in principle, be capable of producing AlphaFold-level breakthroughs across every scientific domain simultaneously.
The Warning Shot
Hassabis's AGI forecast came in the same week that Anthropic's withheld Mythos model — which reportedly discovered thousands of unpatched security vulnerabilities during internal testing — became a focal point for discussions about AI safety. Hassabis referenced Mythos as a 'warning shot': a demonstration that frontier AI systems are already capable of actions with significant real-world consequences, and that the safety frameworks currently in place are not adequate for the systems that will exist in 2029.
The combination of a credible 2029 AGI forecast and a concrete example of a frontier model discovering critical security vulnerabilities creates a specific kind of urgency. It is not the abstract urgency of a distant future risk — it is the urgency of a transition that is close enough to plan for, and that is already producing precursor events that should inform that planning. Whether governments, institutions, and the AI industry itself will respond to that urgency with the seriousness it deserves is the central question of the next three years.