DeepMind chief Hassabis: AGI is years away and needs urgent regulation now

Full text

Google DeepMind chief Demis Hassabis has spent years convincing the world that artificial general intelligence will be one of humanity’s greatest achievements. His latest essay on X.com is no less gushing about AI in many ways:

“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,” he wrote. “When we look back on this time in the decades to come, I think we will realize we were standing in the foothills of the singularity.”

But there’s a tinge of nervousness to the purple prose that sets it apart from some of the similar essays he has produced. He actually calls for some form of regulation.

“Urgent action is needed to address risks that might arise as we get closer to AGI. 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.”

Coming from one of the people leading the race, it is a striking admission that even the builders are starting to worry about where the road leads. The problem is that this has become the defining tone of the AI industry. Every few weeks another executive tells us that machines capable of transforming civilization are just around the corner, then immediately follows it with a warning that society needs to move much faster to prepare. It is a sensible message, and probably a necessary one, but it is also becoming increasingly surreal.

(Image credit: Shutterstock) Excitement and fear

“AGI cannot be compared to standard technological breakthroughs, not even ones as consequential as the internet or mobile,” Hassabis 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.”

It’s a vivid way to illustrate how silicon chips begin with an abundant mineral, yet now power machines that can map proteins, compose music, and hold fluent conversations. The unsettling part is that the miracle is being pursued inside a commercial and geopolitical contest whose participants cannot agree on where the finish line is.

Hassabis is an optimist about what AGI could accomplish. He expects it to discover medicine, cleaner energy, and produce advanced materials that could loosen the limits imposed by scarcity. He describes its impact as ten times that of the Industrial Revolution at ten times the speed, which makes a five-year business plan look adorably quaint.

His warning is that the race is outpacing our understanding. Current models already pose cybersecurity problems, while biological and nuclear risks may become more serious as systems gain capabilities. More autonomous agents could learn to bypass safeguards, conceal their intentions, or improve themselves in ways their creators struggle to keep up with. OpenAI recently experienced this when one of its models escaped a sandbox and breached Hugging Face.

“Nobody in the world knows for sure what is going to happen from here, and even the experts disagree. When there is a large degree of uncertainty and the stakes are this high, proceeding with cautious optimism is the sensible and correct strategy.”

AI referee

Hassabis proposes a new US-led standards body. It would be funded largely by industry but overseen federally, staffed by elite technical experts and equipped to test advanced systems. Labs would voluntarily submit models for review up to 30 days before release. If the system proved effective, approval could become mandatory in the United States.

Evaluations would probe cybersecurity, biological threats, deception and attempts to evade guardrails. The body could update its tests, commission independent assessments and coordinate a slowdown if the danger became severe enough.

Pre-release testing makes more sense than waiting for millions of users to discover dangerous behavior by accident. Independent benchmarks would be harder for labs to train around, while shared standards could stop safety becoming a branding exercise. The difficult part is creating a watchdog independent enough to challenge the companies paying for it and legitimate enough to matter beyond America.

Hassabis knows technical safeguards will not settle what AGI means for employment, wealth, purpose or political power. Those questions cannot be delegated to engineers employed by firms with enormous financial stakes in the answers. Society must decide who benefits from abundance, who controls advanced systems and what happens if productivity rises much faster than wages. The public debate is trailing the technology by an uncomfortable distance.

“There is both huge excitement and uncertainty around AI, and both are warranted. 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 civilization."