Google DeepMind CEO Demis Hassabis issued a public reckoning over Artificial General Intelligence this week, warning that AGI could arrive within the next few years and that the world is not prepared. In a LinkedIn essay on Tuesday, the Nobel laureate compared AGI to the discovery of electricity or fire, arguing its impact could overshadow the Industrial Revolution, and urged governments to write the rules before the technology breaks through.
The post lands as DeepMind and its competitors race to ship AI systems that approach human-level reasoning across most intellectual tasks. Hassabis is no longer framing AGI as a research milestone. He is framing it as a near-term civilizational pivot, and he is asking policymakers, companies, and the public to act accordingly.
Hassabis frames AGI as the next general-purpose technology
Hassabis spent most of his career chasing AGI, and his Tuesday post leaves little doubt that he now believes the chase is nearly over. He wrote that AGI “cannot be compared to standard technological breakthroughs, not even ones as consequential as the internet or mobile,” and added that the technology is “much more akin to the discovery of electricity or fire.” The shorthand he keeps returning to is that humanity has, in his words, “essentially found a way to make sand think.”
That framing matters because two of the three reference points he chose, electricity and fire, are the canonical historical analogies for general-purpose technologies that rewrote the economic and physical order. General-purpose technologies take decades to diffuse, but when they do, they reorganize labor, capital, and geography around entirely new equilibria. If AGI lands anywhere on the same curve, the second half of the 2020s will not look like the first half.
What DeepMind’s CEO is actually worried about
The warning Hassabis attached to that vision is procedural, not technological. He argues that AI is racing ahead of governments and policymakers who are still drafting consultations. There are not, in his telling, enough rules for systems that can match or exceed expert humans across broad cognitive work, and the window to build safeguards before AGI ships is closing.
He does not advocate a pause. He advocates a sequence: AGI should be regulated before it goes mainstream, not after, and the regulatory perimeter should treat advanced models as critical infrastructure rather than as ordinary software products. Once a system can code, plan, and reason across domains, the cost of an alignment failure scales with the breadth of the system itself.
“AGI cannot be compared to standard technological breakthroughs, not even ones as consequential as the internet or mobile. It is much more akin to the discovery of electricity or fire.”
Why the timing is significant
Hassabis published the essay as frontier-model labs enter what insiders call the “capability compression” phase. Capability compression is the period in which successive model generations close the gap to human expert performance across a widening set of tasks, and the gap between demos and deployment shrinks from years to months. The economic consequences in that window are sharper than in either the dot-com boom or the cloud shift, because each new release moves from research preview to enterprise contract faster than the policy cycle can respond.
Regulators in the European Union, the United Kingdom, and parts of the United States have begun writing model-evaluation rules, but most of those rules target current systems, not the AGI-class systems Hassabis is describing. The mismatch is what he is calling out.
What changes if Hassabis is right
- Governments face an unusually short fuse: AGI may arrive before the political cycle that started writing the rules has finished its first major amendment.
- Labor markets absorb a new general-purpose shock on top of the narrow-task automation already in flight.
- Compute supply chains, energy grids, and rare-earth logistics become strategic assets rather than input costs, because AGI workloads scale with the underlying hardware and power envelope.
- Alignment research graduates from an academic subfield into a permanent line item in national security budgets.
What to watch next
Three signals will tell us whether Hassabis’s timeline is structural or aspirational. First, watch whether DeepMind’s next public model release targets an AGI-style benchmark, the way GPT-5 and Claude 4 targeted PhD-level reasoning. Second, watch whether the United Kingdom’s AI Safety Institute expands into a standing regulator with subpoena power, which would mirror the EU AI Act trajectory. Third, watch whether the major cloud providers disclose AGI-scale capacity buildouts in their next quarterly filings, the way hyperscalers disclose AI training cluster expansion today.
Hassabis has staked his public credibility on the claim that the AGI arrival is no longer a research question. It is a calendar question, with all the policy implications that follow. The most important story in artificial intelligence this year is no longer whether the next model is bigger. It is whether the world is ready when it is.