The number everyone gets asked for
p(doom): every public estimate, ranked
p(doom) is shorthand for the probability that advanced AI ends in catastrophe for humanity. 17 people and groups have put a number on it in public, from under 0.01% to 99.999999%. Here they are, lowest first, with a slider so you can see who you agree with.p(doom) is the probability that advanced AI ends in catastrophe for humanity, as guessed by people who should know. 17 public numbers, from under 0.01% to 99.999999%. Most of the people below are still building the thing, which tells you something.
Lowest: under 0.01% Yann LeCun
Highest: 99.999999% Roman Yampolskiy
What p(doom) means
The p is for probability and the doom is for the worst case: human extinction, or a permanent loss of control over the future, depending on who is talking. It started as a half-joke in AI safety circles and became a standard interview question for lab leaders and researchers around 2023. When someone says their p(doom) is 10%, they mean they think there is roughly a one in ten chance that the arrival of advanced AI ends very badly.
It is not a calculation. There is no formula, no agreed definition of doom and usually no time horizon. It is a stated belief, which is why the figures below run from under 0.01% to over 99% among people who have read the same papers. Read it as a measure of how worried someone is, said out loud, with their name attached.
Where do you sit?
Prominent people have put a number on the chance that AI ends badly for humanity. Drag yours into the line-up.
Same as Yoshua Bengio.
- Yann LeCun Chief AI scientist, Meta (at the time), 2023 under 0.01%
- Superforecasters Existential Risk Persuasion Tournament median, extinction by 2100, 2022 0.38%
- AI researchers AI Impacts survey of 2,778 authors, median, 2023 5%
- Vitalik Buterin Ethereum co-founder, 2023 about 10%
- Geoffrey Hinton Turing Award and Nobel laureate, 2024 10 to 20%
- Lina Khan Chair, US Federal Trade Commission (at the time), 2023 15%
- Elon Musk xAI, 2024 10 to 20%
- Dario Amodei CEO, Anthropic, 2023 10 to 25%
- You 20%
- Yoshua Bengio Turing Award laureate, 2023 20%
- Emmett Shear Interim CEO, OpenAI (briefly), 2023 5 to 50%
- Scott Alexander Writer, Astral Codex Ten, 2023 33%
- Paul Christiano Head of safety, US AI Safety Institute, 2023 46%
- Jan Leike Alignment lead, Anthropic, 2023 10 to 90%
- Daniel Kokotajlo AI Futures Project, formerly OpenAI, 2023 70%
- Dan Hendrycks Director, Center for AI Safety, 2023 over 80%
- Eliezer Yudkowsky Machine Intelligence Research Institute, 2023 over 95%
- Roman Yampolskiy AI safety researcher, University of Louisville, 2024 99.999999%
Public statements, placed at the middle of any stated range. Compiled from PauseAI's list and checked 2026-10-30. Ranges and dates matter; follow the link before quoting anyone.
Who said what
15 named people and 2 surveyed groups, lowest first. Each entry is a public statement, placed at the middle of any range the person gave.
-
Yann LeCun: under 0.01%
Chief AI scientist, Meta (at the time), 2023.
More relaxed than Vitalik Buterin (about 10%). Bold.
-
Superforecasters: 0.38%
Existential Risk Persuasion Tournament median, extinction by 2100, 2022.
Same as Yann LeCun.
-
AI researchers: 5%
AI Impacts survey of 2,778 authors, median, 2023.
Between Yann LeCun (under 0.01%) and Vitalik Buterin (about 10%).
-
Vitalik Buterin: about 10%
Ethereum co-founder, 2023.
Between Yann LeCun (under 0.01%) and Geoffrey Hinton (10 to 20%).
-
Geoffrey Hinton: 10 to 20%
Turing Award and Nobel laureate, 2024.
Same as Lina Khan and Elon Musk.
-
Lina Khan: 15%
Chair, US Federal Trade Commission (at the time), 2023.
Same as Geoffrey Hinton and Elon Musk.
-
Elon Musk: 10 to 20%
xAI, 2024.
Same as Geoffrey Hinton and Lina Khan.
-
Dario Amodei: 10 to 25%
CEO, Anthropic, 2023.
Between Elon Musk (10 to 20%) and Yoshua Bengio (20%).
-
Yoshua Bengio: 20%
Turing Award laureate, 2023.
Between Dario Amodei (10 to 25%) and Emmett Shear (5 to 50%).
-
Emmett Shear: 5 to 50%
Interim CEO, OpenAI (briefly), 2023.
Between Yoshua Bengio (20%) and Scott Alexander (33%).
-
Scott Alexander: 33%
Writer, Astral Codex Ten, 2023.
Between Emmett Shear (5 to 50%) and Paul Christiano (46%).
-
Paul Christiano: 46%
Head of safety, US AI Safety Institute, 2023.
Between Scott Alexander (33%) and Jan Leike (10 to 90%).
-
Jan Leike: 10 to 90%
Alignment lead, Anthropic, 2023.
Between Paul Christiano (46%) and Daniel Kokotajlo (70%).
-
Daniel Kokotajlo: 70%
AI Futures Project, formerly OpenAI, 2023.
Between Jan Leike (10 to 90%) and Dan Hendrycks (over 80%).
-
Dan Hendrycks: over 80%
Director, Center for AI Safety, 2023.
Between Daniel Kokotajlo (70%) and Eliezer Yudkowsky (over 95%).
-
Eliezer Yudkowsky: over 95%
Machine Intelligence Research Institute, 2023.
Between Dan Hendrycks (over 80%) and Roman Yampolskiy (99.999999%).
-
Roman Yampolskiy: 99.999999%
AI safety researcher, University of Louisville, 2024.
More worried than Eliezer Yudkowsky (over 95%), which takes some doing.
Why nobody agrees
Three things move the number. What counts as doom: extinction only, or any catastrophe we cannot undo. The horizon: by 2100, or ever. And what is held fixed: the current trajectory, or the world after everyone's best efforts at safety. Two people can share every fact and still be forty points apart because they answered different questions.
The gap does not close with argument. In the Existential Risk Persuasion Tournament, which put superforecasters and AI experts through months of structured debate, the superforecasters finished near 0.38% for AI extinction by 2100 and the experts stayed several times higher. Both groups moved a little and neither moved much.
How this list is kept
Public statements only. Ranges are shown as stated and placed at their midpoint. Corrections welcome as a pull request against this file. Compiled from PauseAI's list and checked 30 October 2026. Ranges and dates matter, so follow a source before quoting anyone. Corrections go as a pull request against the data file.