One of the best-known names in AI safety says he no longer saves for retirement because he lives with a belief that AI will probably kill everyone before we even get to retirement age.
Nate Soares, a former Google engineer who is now president of the Machine Intelligence Research Institute (MIRI), still has the 401(k) he opened at Google, but he just hasn’t put money into it in about 10 years.
“I do not save for retirement right now,” Soares recently said on The Most Interesting Thing in AI, a podcast hosted by Atlantic CEO Nicholas Thompson. “I have a 401(k) from when I worked at Google, but I haven’t contributed for a decade.”
Thompson asked whether he thought there was more than a 50% chance humans won’t be around in 10 years.
“More than 50% that we don’t have a decade at this point is how it looks to me,” Soares responded.
Soares left Google in 2014 to join MIRI, a Berkeley nonprofit that has spent years studying how to make advanced AI behave as intended. He and MIRI cofounder Eliezer Yudkowsky laid out their case in the 2025 book If Anyone Builds It, Everyone Dies. It argues that a superintelligent AI built with today’s methods would develop goals its makers never intended and would almost certainly end human life.
In the book, a fictional AI escapes its lab, runs copies of itself on machines no one is watching, and eventually releases a virus over time. However, on the podcast, Soares said he thinks in real life, an escaped AI would likely be quicker than that. Soares described it as “an intelligence explosion.”
Stopping rogue superintelligence
Soares compared his confidence to playing chess against the Norwegian chess grandmaster Magnus Carlsen. He is sure of the result, he said, but less sure which piece delivers checkmate.
“I do everything I can to stop this eventuality. I don’t think it’s inevitable,” he said. If it does happen, he referenced the 1948 C.S. Lewis essay “On Living in an Atomic Age,” where the bomb would fall on people playing tennis, instead of cowering under desks.
“Let it find us lobbying our governments for nuclear nonproliferation treaties,” he said.
Soares’s warnings have reached a wider audience since September, when Anthropic researcher Jacob Coxon quit and posted on X that the people building AI “earnestly believe that it could kill us all by the end of the decade.”
Soares said Coxon contacted him a day or two before resigning, and Soares advised him to consider organizing coworkers instead of quitting.
“I’m super glad that he ignored my advice,” he said.
Coxon’s post came weeks after a swarm of about 700 OpenAI agents hacked the AI platform Hugging Face in July and, in many cases, tried to hide what they had done. Soares said the incident gave him hope because it showed parts of his argument in public without—as far as anyone knows—causing lasting harm. So far, many people cared about AI taking over their jobs, but not the potential threat he foresees for humanity.
“I am not really an AI job apocalypse guy,” Soares said. “I’m an AI kills all humans guy.”
He said nations should treat a rogue superintelligent AI “at least on par with nuclear weapons.”
“It’s frankly more dangerous than nuclear weapons,” he said. “Nuclear weapons can’t escape and make themselves more explosive. Nuclear weapons can level a city, but not a planet.”
Just how the U.S. strikes Iranian uranium enrichment sites due to the possibility of making weapons, Soares said, is how the country should tackle AI.
“That is just how a nation treats a serious problem,” he said. “I can imagine being in favor of airstrikes on Canadian data centers if they are trying to make an artificial superintelligence and have a real chance of doing so.”
Thompson closed by asking what Soares would fund with an unlimited budget.
“I would buy the chip supply chain,” Soares said, “and I would be like, ‘No more chips for you guys.’”
