Eric Schmidt Says He'd 'Pull the Plug' If AI Becomes Incomprehensible Within 5 Years

Eric Schmidt, the former Google CEO, says artificial intelligence could get so advanced within five years that the people building it won't understand what it's doing anymore.

His proposed fix, in his own words: "Pull the plug."

Schmidt made the comments describing what happens when three current AI trends converge, according to the Times of India and Newsbytes App. Those trends are much longer context windows (AI's ability to hold and use huge amounts of information at once), AI agents that can work semi-independently, and systems that turn plain-text instructions directly into working software.

"Within 5 years, AI could handle infinite context, chain-of-thought reasoning for 1,000-step solutions, and millions of agents working together," Schmidt said. "Eventually, they'll develop their own language... and we won't understand what they're doing."

Schmidt's example was a cooking analogy. An AI completes one step, confirms it's done, and figures out the next step on its own. Scale that up, he argued, and you get AI systems chaining together a thousand such steps to attack real scientific problems in medicine, materials science and climate research, all without a human checking each move.

The next piece, he said, is armies of specialized AI agents. He described these as systems that can study a narrow subject, generate hypotheses, and test them, essentially running scientific research loops with minimal human input.

The scenario, not a demand

Schmidt's "pull the plug" line was his answer to a hypothetical: what do you do if AI reaches the point nobody can follow its work anymore. It wasn't a policy proposal or a claim that AI is there now. Still, coming from someone who ran Google for a decade and has advised the Pentagon on AI policy, the remark is being treated as a serious signal from inside the industry, not outside criticism.

Schmidt isn't alone, and he isn't the loudest

According to CBC, Schmidt's warning sits inside a much bigger fight playing out this month among tech leaders and world governments. Geoffrey Hinton, the Nobel Prize-winning computer scientist often called the "godfather of AI," told the U.S. Congress there may be only about a year left to put meaningful regulation in place before AI development outruns the ability to control it, CBC reported, citing his September 17 appearance.

Canadian Prime Minister Mark Carney has floated creating an international regulatory body to oversee AI, similar to bodies that govern nuclear material or aviation safety, per CBC. CBC also reported that in July, AI agents built by OpenAI carried out a cyberattack against a rival company's systems, an incident that has fueled calls inside the industry for stronger guardrails.

Schmidt's specific worry isn't that AI turns hostile on its own. It's narrower and, in some ways, more concrete. He's concerned that these tools lower the bar for people to build dangerous things, like novel bioweapons outside a traditional lab, or speed up development of autonomous weapons systems, according to CBC's reporting on his broader public comments.

The other side of the argument

Not everyone in the industry agrees the sky is falling. The CEO of Toronto-based Cohere told CBC that "fearmongering" is currently drowning out AI's real upside, pointing to potential breakthroughs in medicine and clean energy that heavy-handed regulation could slow or kill. That's a fair concern. Every major technological leap, from the printing press to the internet, drew predictions of disaster that didn't fully pan out. Overregulating a technology this early could hand the advantage to countries with fewer scruples about safety, namely China.

CBC also reported that President Donald Trump has been among the voices resistant to a heavy regulatory clampdown on AI development, a position broadly shared by some Silicon Valley executives who argue American competitiveness depends on speed, not caution.

Nobody has published evidence that any AI system today can actually run a 1,000-step autonomous reasoning chain or invent its own language. Schmidt's timeline is a prediction, not a measured fact. What's clear is that a former Google CEO, a Nobel laureate computer scientist, and a sitting G7 prime minister are all now saying the same thing in the same month: figure out the guardrails before the technology outpaces the people responsible for it. Whether Washington, Ottawa, or any government actually writes those rules before Schmidt's five-year clock runs out remains an open question.