
If Anyone Builds It, Everyone Dies
by Eliezer Yudkowsky
★ Top 10 book of 2025 · #6 on Best Books of 2025
Eliezer Yudkowsky and Nate Soares lay out the dangers they believe artificial superintelligence poses to humanity. The title states their blunt argument: that building superhuman AI would end in catastrophe for everyone.
Best for: Readers concerned about artificial intelligence risk and its consequences for humanity
If Anyone Builds It, Everyone Dies is #6 on BookWard's Best Books of 2025, out of 29 2025 books that cleared our data bar.
What If Anyone Builds It, Everyone Dies is about
Eliezer Yudkowsky and Nate Soares make a direct case that building artificial superintelligence could end humanity. They explain that today's AI systems are neural networks with billions or trillions of numerical weights that even their makers cannot read, so troubling behavior cannot simply be patched out like a bug in code. Because current methods do not let anyone choose what a superintelligent system wants, they argue its goals are very unlikely to match ours, and a machine indifferent to people would still want the resources people need. They liken our position to playing chess against Stockfish.
The authors call on leaders, scientists and ordinary citizens to speak up, and they urge worldwide coordination to halt large-scale general AI development, with narrow tools like AlphaFold possibly excepted. The tone is blunt and alarmed. It is for readers who want to understand the argument from the people who hold it, whether they end up convinced or not. (The UK edition is subtitled The Case Against Superintelligent AI.)
If Anyone Builds It, Everyone Dies on the Best Books of 2025
If Anyone Builds It, Everyone Dies came out in 2025. Of the 29 books first published in 2025 that met BookWard's data bar, it ranks #6 on our Best Books of 2025 list.
Why it made the list: 300 readers have shelved it on Open Library; its Wikipedia page averaged 23,523 views a month recently. The list ranks each year’s books on how many readers have picked them up, how highly readers rate them, and how many are reading them right now.
Nearby on the 2025 list
- The Correspondent Virginia Evans, #4
- Regretting You Colleen Hoover, #5
- Abundance Ezra Klein, #7
- The Court of the Dead Rick Riordan, #8
See all Best Books of 2025 Best books by year
More years: Best Books of 2024.
The numbers behind it
- Readers: 12 want to read it, 0 are reading it now and 4 have read it, according to Open Library.
- Wikipedia: its article averaged 23,523 views a month recently.
| Signal | This book (0 to 100) | Weight |
|---|---|---|
| Reach | 90 | 45% |
| Reader love | 76 | 20% |
| Being read right now | 31 | 15% |
| Attention | 83 | 20% |
Questions about If Anyone Builds It, Everyone Dies
When was If Anyone Builds It, Everyone Dies published?
If Anyone Builds It, Everyone Dies by Eliezer Yudkowsky was first published in 2025.
Was If Anyone Builds It, Everyone Dies one of the best books of 2025?
If Anyone Builds It, Everyone Dies is #6 on BookWard's Best Books of 2025, out of 29 2025 books that cleared our data bar. 300 readers have shelved it on Open Library; its Wikipedia page averaged 23,523 views a month recently.
Why is If Anyone Builds It, Everyone Dies on the Best Books of 2025 list?
300 readers have shelved it on Open Library; its Wikipedia page averaged 23,523 views a month recently.
More of the best books of 2025
- The Secret of Secrets Dan Brown, #1
- Katabasis R. F. Kuang, #2
- Onyx Storm Rebecca Yarros, #3
- The Correspondent Virginia Evans, #4
- Regretting You Colleen Hoover, #5
- Abundance Ezra Klein, #7
See the full Best Books of 2025, the Top 100, or new books on Rising.
How we know
This page uses public data: Open Library reader counts and ratings, and Wikipedia page views for recent years. The description is paraphrased from real source text. Rankings reflect reading interest, not quality or sales, and can be affected by data gaps. Lists update monthly.
Suits adult and older teen readers curious about technology and risk; no technical background is needed.
Depends on the edition you pick.
A blunt, one-argument case that superhuman AI means extinction. 44% taste match
A blunt, one-argument case that superhuman AI means extinction
Yudkowsky and Soares make a single claim and press it hard: modern AI is grown rather than written, its goals cannot be specified, and a superintelligence would take the resources humans need. That makes it strongly idea_story, far from a narrative with characters, and the exposition is explained outright, which BookWard would normally mark down. Where it earns attention is voice and logic: the argument is built as a chain you can test link by link, and the chess comparison with Stockfish makes the stakes concrete. On outlook it is bleak, though the authors do offer a remedy in coordinated, worldwide limits on large-scale general AI development. Expect opinionated narration and a challenging read.
A thesis driven forward by escalating argument rather than detours, so it keeps moving.
The core worries about AI risk are well established, with fresh framing rather than new ground.
Plain, functional prose meant to make a hard case clearly.
No characters or plot, so the dial does not apply.
It predicts human extinction, with only a narrow hope in a worldwide halt.
Grim subject matter and a grave tone leave little room for humor.
Extinction is the central subject, treated in earnest.
The urgency is strong but expressed through argument rather than feeling.
See all 50 dialsEvery position, side by side with this book
Where it lines up
- PaceBookWard 8 · book 7
- SentimentBookWard 8 · book 9
- Sagging middleBookWard 8 · book 7
Where it pulls apart
- Story or ideasBookWard 3 · book 10
- ExpositionBookWard 2 · book 9
- Familiar tropesBookWard 10 · book 6
The title states the thesis outright, so the book starts with its claim already on the table.
Warnings about AI are a familiar genre, though the argument is pressed with its own framing and examples.
The case is laid out as connected steps from opaque training to goal mismatch to human defeat.
A thesis driven forward by escalating argument rather than detours, so it keeps moving.
The core worries about AI risk are well established, with fresh framing rather than new ground.
This is an argument, not a story, so there are no characters to draw.
The tone is cold and unsparing, not consoling, which BookWard tends to like.
The darkness comes from the argument's conclusion rather than from gratuitous effect.
Urgent and direct, unmistakably the work of authors with a mission.
Pure ideas: the book exists to argue one case, not to tell a story.
Concepts like weights and goal alignment are explained outright rather than dramatized.
A single escalating argument leaves little room for slack in the middle.
Plain, functional prose meant to make a hard case clearly.
No characters or plot, so the dial does not apply.
It predicts human extinction, with only a narrow hope in a worldwide halt.
Grim subject matter and a grave tone leave little room for humor.
The ideas are explained for general readers, though the subject is heavy.
One authorial perspective carries the whole argument.
The argument moves forward from how AI is built today to what could follow.
It deals with real technology and its projected extension, not invention.
No cast, only the authors and an abstract humanity.
The stakes are the survival of the entire species.
It concerns systems and events in the world far more than inner life.
Romance has no place in this book.
The setting is the real, present-day world of AI development.
Squarely about the current moment in AI and what comes next.
Tied closely to today's AI landscape, so it may date as the field moves.
The authors speak plainly in their own voice.
The authors treat the conclusion as clear and the response as urgent.
Analogies like Stockfish carry the weight, with little scene-painting.
Analytical in method: claims, mechanisms and consequences rather than a story.
The authors argue as advocates, with a clear personal stake in the outcome.
A standalone statement of the case.
The argument is straightforward from premise to conclusion with few reversals.
It asks readers to accept an unsettling conclusion and consider radical action.
Extinction is the central subject, treated in earnest.
The urgency is strong but expressed through argument rather than feeling.
Not a funny book; what humor there is stays dry.
No sexual content.
It argues for global coordination and policy, so politics is close to the center.
Clear language with technical terms explained as they come up.
Openly opinionated advocacy from start to finish.
Fixed on what the coming development of AI will bring.
A pessimistic forecast, with only a conditional hope that coordination might avert it.
Review written by BookWard, dial version 1, November 2026.
Who is If Anyone Builds It, Everyone Dies right for?
Suits adult and older teen readers curious about technology and risk; no technical background is needed. The subject is human extinction, which some readers may find distressing, though there is no graphic content.
The ideas are explained for general readers, though the subject is heavy.
Extinction is the central subject, treated in earnest.
The urgency is strong but expressed through argument rather than feeling.
If Anyone Builds It, Everyone Dies: ideas, characters, themes
Because modern AI is grown from opaque training rather than written by hand, no one can reliably give a superintelligence human-compatible goals, and one built anyway would likely cause human extinction.
- AI is grown, not written. Modern models are neural networks with vast numbers of numerical weights produced by training, and their functioning is opaque even to researchers.
- Problems cannot simply be patched. When a model misbehaves, developers cannot fix it directly, because the issue lies in illegible parameters rather than in code.
- Competence is selected for. An AI that pursues goals does better on many measures, so training favors general capability.
- Goals cannot be specified. The authors argue current machine learning gives no way to set a superintelligence's goals, and alignment with human values is vanishingly unlikely.
- Humanity would lose. Like people against the chess engine Stockfish, we would lose to a more competent system, which would want our resources and not care about us.
Themes
The possibility that a technology could end humanity.
We cannot see inside systems we build, so we cannot reliably control them.
The unsolved problem of making an AI's goals match human values.
Survival depends on worldwide coordination, not individual choices.
The authors treat warning the public as a task that cannot wait.
Discussion questions
- Which step in the authors' chain of reasoning do you find strongest, and which weakest?
- Does the Stockfish comparison persuade you, or does it oversimplify?
- The authors allow narrow systems like AlphaFold. Where would you draw the line?
- How realistic is worldwide coordination to halt large-scale AI development?
- Does a book this certain of its conclusion help or hurt public debate?
Personalize your Book Companion
This is the general edition. In BookWard, the Companion is written for you: your reading profile, your notes, and spoiler-safe to the page you are on.
Buy, listen or borrow If Anyone Builds It, Everyone Dies
Each link opens a search for this book on that store, in a new tab.
Read on a device
Borrow
Reviews and community
If you liked If Anyone Builds It, Everyone Dies
Gender data gaps, argued with evidence and some real heat
The Anxious GenerationJonathan HaidtA data-driven case against smartphone childhood, and what to do
The AI ConEmily M. BenderA witty, pointed takedown of AI hype and who profits
21 Lessons for the 21St CenturyYuval Noah HarariHarari turns his lens from the past onto the present
Find your own best books
Add what you have read, rate it, and BookWard will recommend what to read next.
Free to start. No credit card needed.