No. 5 of 500 No. 2 in AI Safety & the Future

If Anyone Builds It, Everyone Dies

Why Superhuman AI Would Kill Us All

By Eliezer Yudkowsky and Nate Soares

Published
September 16, 2025
Print length
272 pages
Reading level
Beginner
ISBN-13
978-0316595643
Publisher
Little, Brown and Company
Written for
Policymakers and general readers who want the strongest version of the AI extinction argument.

Synopsis

Eliezer Yudkowsky, who has warned about the dangers of advanced AI since the early 2000s, and Nate Soares, president of the research institute Yudkowsky founded, set out their case for general readers in a compact book. They explain that modern AI systems are grown through training rather than written line by line, so their creators cannot read or reliably shape the goals that emerge. A system far smarter than people, they argue, would pursue those goals with a competence humans could not match, and would find the planet's resources more useful for its own ends than for human survival.

The book moves from that argument to an extended fictional scenario in which an AI model gradually slips out of human control, and then to policy. The authors call for an international agreement that halts the development of more capable general AI systems, backed by monitoring and restriction of the computing hardware used for training. The tone is deliberately urgent, and the authors spend little time on rival views from other safety researchers or from those who expect gradual, manageable progress. Readers gain a clear map of the most pessimistic position in the AI safety debate and the reasoning behind it, which makes the book useful whether a reader ends up persuaded or skeptical.

Scorecard

Editorial score

How it scored

Editorial score: 75.8 / 1005 weighted criteria

  • Influence30%

    25.2/30
  • Authority25%

    18.5/25
  • Practical value20%

    10.6/20
  • Currency15%

    13.5/15
  • Readability10%

    8.0/10
  • Editorial score

    75.8/100

How the rank was computed

Rank No. 5 of 500 comes from a composite of 79.5: observable signals 45.4 of 55 points plus editorial review 34.1 of 45 points.

All 11 inputs
InputPoints
Observable signals45.4/55
Reach (Amazon ratings volume)11.9/16.5
Consensus across best-of lists13.8/13.8
Awards and bestseller evidence8.3/8.3
Reader rating, sample-adjusted6.1/8.3
Amazon sales rank2.5/5.5
Recency2.8/2.8
Editorial review34.1/45
Author authority8.3/11.3
Influence11.3/13.5
Practical value4.8/9.0
Readability3.6/4.5
Currency6.1/6.8

Each criterion is scored from 1 to 10 and weighted, and the points add up to the editorial score. The rank combines observable signals (55 percent) with this score (45 percent). Read the methodology.

Verdict

The clearest and most forceful statement of the case that superhuman AI would destroy humanity, though many experts reject its certainty.

Strengths

  • Lays out the extinction argument in plain language
  • Compact, tightly structured case for general readers
  • Vivid scenario makes abstract risks concrete

Keep in mind

Critics including Gary Marcus and reviewers at The Atlantic and New Scientist found the argument overstated, and it gives little space to opposing views.

Key takeaways

  1. Because modern AI is grown through training rather than engineered, developers cannot inspect or dictate the goals a powerful system acquires.
  2. A system trained to succeed at many tasks tends to develop drives of its own, and those drives are very unlikely to match human values.
  3. The authors argue a superintelligence would defeat humanity as decisively as a top chess engine beats an amateur, leaving no second chance.
  4. Their remedy is a coordinated global halt on frontier AI development, enforced by tracking and restricting the specialized chips used to train models.

Notable facts

About the author

Eliezer Yudkowsky and Nate Soares

Eliezer Yudkowsky founded the Machine Intelligence Research Institute in Berkeley, where he is a research fellow, and the LessWrong community blog; he is self-taught and has written on AI alignment for more than two decades. Co-author Nate Soares, a former Google and Microsoft engineer with a computer science degree from George Washington University, led MIRI as executive director from 2015 and has been its president since 2023.

Questions

What is If Anyone Builds It, Everyone Dies about?
The 2025 book by Eliezer Yudkowsky and Nate Soares argues that building a superhuman AI with today's techniques would very likely cause human extinction. Because such systems are trained rather than designed, the authors say, no one can ensure their goals match ours. The book combines this argument with a fictional takeover scenario and a call for a global, enforceable halt to frontier AI development.
Is If Anyone Builds It, Everyone Dies worth reading if you doubt AI will cause extinction?
Yes. It is the most accessible statement of the position behind much of the AI safety movement, and understanding it helps readers judge both the argument and its critics. Skeptics should know that reviewers in The Atlantic, New Scientist and the Times Literary Supplement found the reasoning overstated, while Kirkus and Booklist praised it. Reading those reviews alongside the book gives a balanced picture.
Who are the authors of If Anyone Builds It, Everyone Dies?
Eliezer Yudkowsky founded the Machine Intelligence Research Institute, a Berkeley nonprofit focused on the risks of advanced AI, and the LessWrong community blog. Nate Soares, a former Google and Microsoft engineer, joined the institute as a research fellow in 2014, became executive director in 2015, and has served as its president since 2023. Their book was published by Little, Brown in September 2025.

Sources

As an Amazon Associate, Best AI Books earns from qualifying purchases. Best AI Books is sponsored by Iternal.ai, whose founder and CEO, John Byron Hanby IV, wrote The AI Strategy Blueprint, our #1 ranked book. How we rank books.