If Anyone Builds It, Everyone Dies

Eliezer Yudkowsky, co-founder of the Machine Intelligence Research Institute (MIRI), and Nate Soares, its president, argue that the race to build artificial superintelligence is one humanity cannot survive. If Anyone Builds It, Everyone Dies sets out their case that AI companies are building systems they do not know how to control, and that nobody yet knows how to make a mind far smarter than ours reliably want what its creators want.

The authors explain that modern AI models are vast networks of numerical weights that even their developers cannot read, so a behaviour nobody wanted cannot simply be fixed, and that training selects for capable, goal-pursuing systems whose actual goals nobody chose. Such a system, they argue, would have no reason to keep people around, and a contest against a mind that outthinks us at everything is one humanity would lose, just as people lose at chess to the engine Stockfish.

The middle of the book is a fictional scenario in which an AI company called Galvanic builds a system named Sable, tracing step by step one way such a system could slip out of human control. Later chapters describe the state of the field as "an alchemy, not a science" and close with a policy argument: that governments should coordinate, through international treaties and limits on the computing hardware used to train AI, to halt large-scale development of general AI everywhere.

Published in September 2025, the book became a New York Times bestseller and appeared on The New Yorker's best books of the year so far and The Guardian's best science and nature books of 2025. Publishers Weekly called it "a frightening warning that deserves to be reckoned with," while noting that it offers few opposing viewpoints.