P(doom) is the probability that AI causes human extinction or a permanent catastrophe. This calculator builds yours from five step-by-step questions, then compares it with 19 experts, from Yann LeCun at under 0.01% to Roman Yampolskiy at 99.9%, and surveys where the median AI researcher says 5%.
Answer five questions with a percentage. Each one is conditional on the step before it. Nothing leaves your browser.
Step 1 · AGI is built by 2100
AI that can do nearly every cognitive job as well as the best humans. Forecasts for when this arrives range from the late 2020s to after 2100.
Step 2 · Alignment fails
At least some powerful AI systems end up pursuing goals their developers did not intend, because nobody has solved how to reliably steer them in time.
Step 3 · It is deployed anyway
Competition pushes it into use and it acts to gain resources or resist shutdown in high-impact ways. A pause, regulation or lab caution that stops this lowers the number.
Step 4 · We can't stop it
Nobody detects, corrects or contains it before it kills or permanently disempowers humanity. Warning shots, monitoring and other AI systems on our side lower the number.
Second path · Catastrophic misuse
A separate path where the AI does what it is told: an engineered pandemic, an AI-driven war or a permanent totalitarian lock-in.
Log scale from 0.01% to 100%, so each step is 10 times the last. The dark line is you.
Superforecasters (XPT, 2022) · 0.38%
Median chance of AI-caused human extinction by 2100 from 89 superforecasters with strong short-term track records.
AI experts (XPT, 2022) · 3%
Median chance of AI-caused human extinction by 2100 from AI domain experts in the same tournament, about 8 times the superforecasters.
AI researchers (2023 survey) · 5%
Median chance that human-level AI leads to "extremely bad (e.g. human extinction)" outcomes, from 2,778 authors of top AI papers. The mean was 9%.
Marc Andreessen · 0%
Co-founder, Andreessen Horowitz · said Mar 2024
Added "p(Doom) = 0" to his X bio after calling AI risk a "moral panic". Source for Marc Andreessen (opens in a new tab)
Yann LeCun · <0.01%
Turing Award winner, then Meta's chief AI scientist · said Apr 2024
Puts the chance at under 0.01%: less likely than an asteroid wiping us out. Source for Yann LeCun (opens in a new tab)
Nate Silver · 5-10%
Statistician, Silver Bulletin · said Jan 2025
"My estimate of p(doom) would be considered high by normal person standards but is well in line with the expert consensus, which puts the chances at 5 to 10 percent." Source for Nate Silver (opens in a new tab)
Lex Fridman · 10%
Podcast host and AI researcher · said Jun 2025
"I would say my p(doom) is about 10%." Source for Lex Fridman (opens in a new tab)
Toby Ord · 10%
Oxford philosopher, author of The Precipice · said Mar 2020
Puts the chance of existential catastrophe from unaligned AI this century at one in 10. Source for Toby Ord (opens in a new tab)
Joe Carlsmith · >10%
Philosopher, author of the power-seeking AI report · said May 2022
Estimated ~5% for an existential catastrophe from power-seeking AI by 2070 in 2021, then raised it to over 10%. Source for Joe Carlsmith (opens in a new tab)
Geoffrey Hinton · 10-20%
Turing Award and Nobel Prize winner · said Dec 2024
Says there is a 10% to 20% chance AI leads to human extinction within the next three decades. Source for Geoffrey Hinton (opens in a new tab)
Elon Musk · 10-20%
CEO of Tesla and SpaceX, founder of xAI · said Mar 2024
"I probably agree with Geoff Hinton that it's about 10% or 20% or something like that." Source for Elon Musk (opens in a new tab)
Lina Khan · 15%
Then chair of the US Federal Trade Commission · said Dec 2023
Told The New York Times' Kevin Roose she is at 15 percent. Source for Lina Khan (opens in a new tab)
Dario Amodei · 10-25%
CEO of Anthropic · said Dec 2023
Puts his p(doom) between 10 and 25 percent. In September 2025 he said there is a 25% chance things go "really, really badly". Source for Dario Amodei (opens in a new tab)
Yoshua Bengio · 20%
Turing Award winner, founder of Mila · said Jul 2023
"I got around, like, 20 per cent probability that it turns out catastrophic." Source for Yoshua Bengio (opens in a new tab)
Emmett Shear · 5-50%
Twitch co-founder, briefly OpenAI's interim CEO · said Dec 2023
Has said he hovers somewhere between 5 and 50 percent. Source for Emmett Shear (opens in a new tab)
Scott Alexander · 33%
Writer, Astral Codex Ten · said Mar 2023
"If you force me to give an estimate it's probably around 33%; I think it's very plausible that we die, but more likely that we survive." Source for Scott Alexander (opens in a new tab)
Paul Christiano · 46%
Alignment researcher, ex-OpenAI, founder of ARC · said Apr 2023
46% that humanity irreversibly messes up its future within 10 years of building powerful AI, including 22% for an AI takeover. Source for Paul Christiano (opens in a new tab)
Holden Karnofsky · 10-90%
Co-founder of Open Philanthropy · said Aug 2022
Has little sympathy for anyone who puts the odds of AI takeover "under 10% or over 90%". Source for Holden Karnofsky (opens in a new tab)
Daniel Kokotajlo · 70%
Ex-OpenAI researcher, lead author of AI 2027 · said May 2023
"I think AI doom is 70% likely and I think people who think it is less than, say, 20% are being very unreasonable." Source for Daniel Kokotajlo (opens in a new tab)
Dan Hendrycks · >80%
Director, Center for AI Safety · said Dec 2023
Puts his p(doom) above 80%, up from about 20% two years earlier. Source for Dan Hendrycks (opens in a new tab)
Eliezer Yudkowsky · >95%
Founder of MIRI, co-author of If Anyone Builds It, Everyone Dies · said Dec 2023
Puts the chance of doom above 95%. Source for Eliezer Yudkowsky (opens in a new tab)
Roman Yampolskiy · 99.9%
AI safety researcher, University of Louisville · said Jun 2024
Predicts a 99.9% chance that AI ends humanity within the next 100 years. Source for Roman Yampolskiy (opens in a new tab)
Ranges are plotted at their midpoint; 0% and <0.01% sit on the left edge. People mean different things by "doom" and use different time frames, so read each quote. Checked against sources November 13, 2026.
Give a percentage for AGI by 2100, alignment failing, deployment anyway, failing to stop it and catastrophic misuse.
See your P(doom), how much comes from AI takeover vs misuse, and what it means out of 1,000 futures.
See where you sit against 19 named experts and 3 surveys on a log scale, each with a quote and source.
Copy the link or a text summary. Your answers are saved in the link.
The takeover chain follows the argument philosopher Joe Carlsmith laid out in his 2021 report Is Power-Seeking AI an Existential Risk?: powerful AI gets built, aligning it is hard, misaligned systems get deployed and seek power anyway, and that escalates to humanity losing control. Multiplying his estimates for each step gave him about 5%, later over 10%. Yoshua Bengio built his 20% the same way. Misuse is a second path, because an AI that does exactly what it is told can still be pointed at a catastrophe. Splitting the number into steps shows which part of the argument you actually disagree with.
P(doom) is shorthand for the probability that AI leads to human extinction or another permanent catastrophe, such as humanity losing control of its future for good. It began as an in-joke on the LessWrong forum and spread in 2023 when AI researchers and tech leaders started stating their own numbers in public.
In the 2023 survey of 2,778 AI researchers, the median chance that human-level AI leads to extremely bad outcomes such as human extinction was 5% and the mean was 9%. In the 2022 Existential Risk Persuasion Tournament, AI experts gave a median 3% chance of AI-caused extinction by 2100 and superforecasters 0.38%. Individual estimates run from 0% to 99.9%.
It multiplies a chain of conditional chances: AGI is built by 2100, alignment fails, the misaligned AI is deployed and seeks power anyway, and humanity fails to stop it. It then adds a separate path for people misusing AGI to cause a catastrophe, without double counting. The formula is P(AGI) x [1 - (1 - alignment fails x deployed x can't stop it) x (1 - misuse)].
In December 2024 Geoffrey Hinton said there is a 10% to 20% chance AI leads to human extinction within the next 30 years. Elon Musk has said he agrees with Hinton at about 10% or 20%, Dario Amodei has put his between 10% and 25%, and Yann LeCun puts it below 0.01%.
Only roughly. People rarely say what counts as doom or by when, so a 10% from one person and a 10% from another can mean different things. It is most useful for showing where someone stands and which step of the argument they doubt, which is why this calculator breaks it into parts.
Estimates were checked against their sources on November 13, 2026. They are personal opinions about an uncertain future, not measurements, and people may have changed their numbers since. AI Wins is not affiliated with any person or organization listed.
When will AGI arrive? Compare AI 2027 and expert forecasts.
Open toolIs AI a bubble? Score 10 warning signs vs the dot-com era.
Open toolSee how exposed your job is to AI automation.
Open toolPlain-English definitions of AGI, alignment and more.
Open toolTrack the AI safety news that moves these numbers in the free daily AI Wins digest.
Get the free digestOnly positive AI stories, delivered daily or weekly. Free, no spam, unsubscribe anytime.