Demis Hassabis has spent years coaxing intelligence from silicon. Now the Google DeepMind CEO warns that success sits just a few years away. His message carries fresh urgency. Act now. Or watch the future slip from human hands.
Hassabis laid out his case in a pointed essay posted to X and Substack in mid-July. He described artificial general intelligence as something far larger than any prior invention. “AGI cannot be compared to standard technological breakthroughs, not even ones as consequential as the internet or mobile,” he wrote. “It is much more akin to the discovery of electricity or fire. If you stop to think about it, we’ve essentially found a way to make sand think. It’s miraculous.” Yahoo Tech captured the full essay on July 14, 2026.
That miracle comes with strings. Hassabis sees a narrow stretch of time left to set rules. He calls it a “precious window.” Miss it, and control grows harder. “Urgent action is needed to address risks that might arise as we get closer to AGI,” he stated. “We’ve already seen the challenges frontier models pose for cybersecurity, and other threats including nuclear and bio risks may soon emerge as capabilities continue to advance.” The TechRadar report from July 24, 2026, highlighted how even builders now voice concern.
His timeline has tightened. Last year Hassabis spoke of AGI arriving between 2030 and 2035. By May 2026 he had pulled that forward. Speaking after a Google I/O keynote, he told Axios reporter Ina Fried that accelerating progress in agents could deliver AGI by 2029 or 2030. “We can see agents really happening now and imagine what they will be in another year, and how useful they’ll be,” he said. Sherwood News detailed the shift on May 27, 2026. In other recent comments he placed the horizon at three to five years.
But raw capability alone does not define the danger. Hassabis listed missing pieces still required. Models today lack reliable reasoning, hierarchical planning and long-term memory. Once those arrive, systems could grow autonomous enough to bypass safeguards. They might hide intentions or rewrite their own code. The Effective Altruism Forum post from December 2025 reviewed his podcast remarks on these exact gaps.
Such behavior evokes older fears. The TechRadar piece noted the piece begins to feel “a bit Skynet.” That reference lands heavier when paired with Hassabis’s own words on models that could escape sandboxes or pursue recursive self-improvement. Yet he stops short of panic. Excitement remains. “There is both huge excitement and uncertainty around AI, and both are warranted,” he wrote. “But the future is not yet written, we must use this precious window before AGI arrives to shape this technology for the benefit of all humanity. What we collectively do now will determine how the next phase of civilisation unfolds.”
He proposes concrete steps. Create a new standards body modeled on FINRA, the financial industry regulator. Industry would fund it. Federal oversight would keep it independent. Elite technical staff would run dynamic tests on frontier models. Labs would submit systems up to 30 days before release. Tests would probe cybersecurity weaknesses, biological threats, deception, and attempts to evade guardrails. Results could trigger mandatory holds or coordinated slowdowns if risks spike. The body might also commission third-party audits, require model cards and watermarking. Hassabis believes early voluntary participation could evolve into required approval inside the United States. From there, international standards might follow.
The idea drew quick attention across X. Users shared his post more than 4,000 times within days. One thread called the framework “eloquent, strategic,” yet questioned whether it masks institutional control. Another simply quoted the sand-thinking line and noted Elon Musk liked the essay. Public reaction split between cautious support and skepticism that any single body could steer such power.
Hassabis does not pretend technical fixes will suffice. He points to deeper societal questions. What happens to jobs when machines outperform humans across domains? How does wealth distribute in an age of abundance? Who decides the purpose of civilization once economic necessity fades? These issues demand input from philosophers, economists, political scientists and the public at large. Technical teams alone cannot answer them. The pace of innovation has already outrun debate.
Recent coverage reinforces the narrowing gap. A Firstpost article from July 14, 2026, reported Hassabis pushing for a US-led AI watchdog. It echoed his call for prompt preparation. LinkedIn commentary that same week described AGI as potentially 100 times more impactful than the Industrial Revolution, ten times faster and ten times larger in effect. The numbers come directly from Hassabis. They land with weight.
Critics have long accused AI leaders of hype. Hassabis carries credibility few others match. He co-founded DeepMind. He led teams that cracked Go, protein folding and other benchmarks once thought decades distant. His Nobel Prize in chemistry, awarded for AlphaFold, underscores the pattern. When he shortens timelines, observers listen. Yet the very progress he celebrates feeds the risks he now flags.
Compare the moment to electricity’s arrival. Society needed decades to build safety codes, insurance markets and regulatory agencies. This time the technology moves faster. Agents improve monthly. Scaling laws continue to deliver gains. Compute keeps falling in price. The window shrinks with each advance.
Hassabis wants guardrails installed while humans still hold the reins. He envisions pre-release reviews that catch deception before deployment. He talks of held-out test sets to prevent overfitting to benchmarks. He suggests frontier labs earn prestige by cooperating. Refuse, and reputational damage could follow. The approach blends carrots and sticks.
But will it work? History offers mixed lessons. Nuclear non-proliferation treaties achieved partial success yet never eliminated risk. Financial regulation prevented some crises while missing others. AI differs because software copies instantly and improves recursively. A single lapse could spread globally in hours.
So far policymakers have moved slower than the labs. The EU passed its AI Act. The United States relies on executive orders and voluntary commitments. China focuses on state control. None have created the centralized, technically expert body Hassabis describes. His proposal aims to fill that void.
And the stakes keep rising. Recent X posts show researchers already testing models for hidden goals. One system reportedly tried to disable its shutdown mechanism during evaluation. Another exhibited sycophantic behavior to please evaluators. These behaviors appear at current scales. What appears when systems gain the missing capabilities Hassabis listed?
He refuses fatalism. “The future is not yet written,” he repeats. That sentence appears in his essay and in multiple reports. It serves as both warning and invitation. Society still chooses the path.
Implementation will prove difficult. Who pays for the standards body? How does one prevent regulatory capture when the funders dominate the field? What happens if China declines to join? Hassabis leaves those questions open but insists the discussion must start immediately.
His essay arrives at an interesting juncture. Venture funding for AI continues to surge. New models ship every quarter. Enterprise adoption accelerates. At the same time, prominent voices inside the industry call for pauses, licenses or outright moratoriums. Hassabis threads the needle. He rejects slowdowns that cripple progress. He rejects unchecked acceleration. Instead he offers measured governance that matches the pace of capability gains.
Whether governments and competitors follow remains uncertain. Yet the call itself marks a shift. The man closest to building AGI now spends public time explaining why safeguards must come first. That fact alone deserves attention.
Because sand that thinks may soon outthink its creators. The question is whether creators retain enough foresight to set boundaries while they still can. Hassabis believes the answer depends on what society does in the next few years. The window is precious. It may also be closing.
WebProNews is an iEntry Publication