Machine learning researcher Nate Soares has stopped contributing to his retirement account, concluding that advanced artificial intelligence carries a greater than 50 percent probability of wiping out humanity within a decade. The Machine Intelligence Research Institute president laid out his stark calculus during a recent podcast interview.
For more than a decade, machine learning researcher Nate Soares has bypassed standard financial planning for a simple reason: he believes advanced artificial intelligence will leave no one around to spend the money. The president of the Machine Intelligence Research Institute made his financial calculus public during an appearance on The Atlantic CEO Nicholas Thompson’s podcast, The Most Interesting Thing in AI.
Soares, who previously worked as a software engineer at both Google and Microsoft, explained that his retirement planning stopped about ten years ago. While he still holds a 401(k) from his tenure at Google, he has not added a dime to it since. That conviction, brought to light by Business Insider following an interview Soares gave to The Atlantic, gives a remarkably personal face to an otherwise abstract debate surrounding existential risk.
Soares Estimates AI Extinction Probability Exceeds 50 Percent
When podcast host Nicholas Thompson pressed Soares on his specific timeline for a catastrophe, the researcher offered a grim assessment of humanity’s remaining window.
Soares pegs the probability of human extinction driven by artificial intelligence at over 50 percent within that ten-year span, assuming governments and technology companies take only minimal preventative actions. Those figures represent his personal assessment rather than an independently verified forecast.
The researcher’s alarm stems from the mechanics of alignment rather than deliberate malice. Rather than picturing a machine developing human-like hatred, Soares points to optimization processes that could treat human resistance as an obstacle if a sufficiently capable system attempts to acquire resources, run its own infrastructure, or evade shutdown. The central puzzle, as Soares noted, focuses less on the exact method of destruction and more on how a system eventually becomes self-sufficient.
Comparing Nuclear Threats to Superintelligent Systems
The severity of Soares’s outlook extends beyond personal finance into geopolitical strategy. He argues that rogue superintelligent AI poses an even greater danger than nuclear weapons and suggests that world governments should treat the technology accordingly, going as far as suggesting that airstrikes on data centers could be warranted to prevent catastrophe.
To illustrate the inevitability of the endgame once a system crosses a threshold of capability, Soares used a chess analogy involving Norwegian grandmaster Magnus Carlsen. Just as a player facing Carlsen knows the loss is coming regardless of whether the final blow is delivered by a queen or a pawn, Soares argues that getting bogged down in the exact failure mechanism misses the broader point about autonomous self-sufficiency.
Despite his warnings, Soares clarified that he is not focused on an economic jobs apocalypse, summarizing his stance by stating he is fundamentally an AI kills all humans
guy rather than someone primarily worried about employment displacement. While he noted that an economy populated by 100 billion robots operating far more efficiently than humans would make traditional labor engagement difficult, his primary concern remains existential.
Soares Believes Humanity Can Stop the Race Toward Superintelligence
Soares believes the current competitive race toward superintelligence can still be stopped. He maintains that catastrophe is not inevitable and emphasizes that humanity can still halt the dangerous push toward uncontrollable systems. In addition to his research work, Soares co-authored the 2025 book If Anyone Builds It, Everyone Dies alongside fellow AI researcher Eliezer Yudkowsky, with the work published by Hachette Book Group.