Artificial General Intelligence (AGI), often described as AI that can match or even surpass human-level cognitive abilities, may no longer be a distant concept. As per Google DeepMind CEO Demis Hassabis, the technology could arrive “within a few short years.” But alongside that prediction, the AI pioneer is also calling for stronger safeguards. In a new essay, Hassabis argues that the world’s most powerful AI systems should undergo independent safety testing before they are released to the public. Here’s everything you need to know.
AGI Could Be As Transformative As Electricity
In an essay titled A Framework for Frontier AI and the Dawning of a New Age, shared on X, Hassabis describes AGI as one of the biggest technological shifts humanity has ever faced.
“This is a pivotal moment in human history. Artificial General Intelligence (AGI), a system that exhibits all the cognitive capabilities the brain has, is probably only a few short years away.”
Rather than comparing AGI to the internet or smartphones, Hassabis believes its impact could rival breakthroughs such as electricity and fire. If developed responsibly, he says the technology could dramatically accelerate drug discovery, enable cleaner energy solutions, unlock new materials, and boost productivity across nearly every industry.
He also offered one of the essay’s most memorable lines: “Humans have essentially found a way to make sand think.”
The Google DeepMind CEO says, AGI could deliver an impact “10x the Industrial Revolution at 10x the speed.” At the same time, he acknowledges that such progress also brings serious risks. He warns that future frontier AI systems could amplify cybersecurity threats and even create biological or nuclear security concerns if left unchecked.
A New Standards Body To Test Powerful AI Models
To prepare for the AGI era, Hassabis proposes establishing a US-led Frontier AI Standards Body that would independently evaluate advanced AI models before public deployment.
Under his proposal, companies developing highly capable AI systems would voluntarily submit their models for testing up to 30 days before launch. These evaluations would examine cybersecurity risks, biological threats, deceptive behaviour, and whether models can bypass built-in safety guardrails. Developers would also be encouraged to strengthen cybersecurity, publish detailed model cards, and invest more heavily in AI safety research.
If the framework proves effective, Hassabis believes these assessments could eventually become mandatory for frontier AI models released in the US before expanding into a broader international standard.
“The field is now locked in an extremely intense commercial and geopolitical race,” he writes, adding that AI capabilities are advancing faster than society’s understanding of them. “Nobody in the world knows for sure what is going to happen from here,”
Hassabis argues that “cautious optimism” is the right way to approach the coming AGI era, adding that the choices made today will ultimately determine how the technology shapes the future of humanity.
