Explainer
What is p(doom)?
p(doom) is shorthand for “probability of doom”: one person’s estimate, as a percentage, that advanced AI leads to a catastrophic outcome for humanity. It is an opinion expressed as a number. There is no agreed method for producing it and no way to check one.
“At some stage therefore we should have to expect the machines to take control.”ALAN TURING, 1951 · “INTELLIGENT MACHINERY, A HERETICAL THEORY”
Where do 28 named p(doom) estimates sit on one ladder?
Each bar is one person’s publicly stated p(doom), with where they said it. A range means they gave a range. The dashed line is the survey median (5% of 2,778 researchers); the middle voice here says 27.5%. Figures as compiled 2026-03; people revise them.
Drag to place yourself among the voices.
On this dial 100% is midnight, so 51 minutes to go. That clock is your opinion. SIREN’s own reading, level 4 right now, isn’t an opinion: it counts what’s happening in AI every hour, with receipts.
What is in the news about AI catastrophe, safety, and p(doom)?
- Superintelligence Is Now a Product Label. Here Is What Builders Should Do With Itgnews-ai · 36 min ago
- At Anthropic pre-IPO meet: ‘Thinking’ caps, surging revenue, no human extinction By Reutersgnews-ai · 48 min ago
- NYU Professor Gary Marcus on AI Safetybloomberg-tech · 1 h ago
- Trump: US leads world in superintelligencegnews-ai · 3 h ago
- Stavian discusses superintelligence infrastructure with RAND experts.gnews-ai · 3 h ago
- AI Safety Is Not A Corporate Plotreddit-controlproblem · 4 h ago
Refreshes live from SIREN’s newsroom. Which AI stories has SIREN scored in this window? → · What AI is on TV and radio right now? → · What happens the day after an AI catastrophe? →
Which polls and surveys report p(doom)?
- AI researchers: 5% median, 9% mean. 2,778 authors of top AI papers asked about an outcome as bad as human extinction. AI Impacts / Grace et al., 2024
- Harvard event attendees: 50% → 70% median. 89 people, before and after a talk on “If Anyone Builds It, Everyone Dies”; self-described experts didn’t move up. Kestin & Soares, arXiv, Mar 2026
What further reading covers every side?
- Does AI pose an existential risk? We asked 5 expertsThe Conversation · Jul 2026
- The peril and opportunity of artificial superintelligenceNPR, 1A · Jun 2026
- Decoding AI existential riskCNN · Jun 2026
- Power-seeking superintelligence: possible, but improbable in the short termAI and Ethics · 2025
- If Anyone Builds It, Everyone DiesYudkowsky & Soares · 2025
- Stephen Hawking warns AI could end mankindBBC · Dec 2014
- What AI books argue both sides of the case? →
Where each figure comes from
- Grady Booch: 0% · Fast Company, Jul 2023
- Marc Andreessen: 0% · The New Yorker, Mar 2024
- Yann LeCun: 0.01% · TechRadar, Apr 2024
- Benjamin Mann: 0–10% · Lenny’s Podcast, Jul 2025
- Casey Newton: 5% · Fast Company, Jul 2023
- Nate Silver: 5–10% · Silver Bulletin, Jan 2025
- Lex Fridman: 10% · Lex Fridman transcript, Jun 2025
- Toby Ord: 10% · ABC News, Oct 2023
- Vitalik Buterin: 12% · Doom Debates, Aug 2025
- Geoffrey Hinton: 10–20% · METR Q&A, Jun 2024
- Lina Khan: 15% · New York Times, Dec 2023
- Dario Amodei: 10–25% · Axios, Sep 2025
- Elon Musk: 10–30% · Business Insider, 2024
- Emmett Shear: 5–50% · New York Times, Dec 2023
- Shane Legg: 5–50% · LessWrong Q&A, 2011
- Emad Mostaque: 50% · Post on X, Dec 2024
- Holden Karnofsky: 50% · The Spectator, Mar 2024
- Jan Leike: 10–90% · Fast Company, Jul 2023
- Paul Christiano: 50% · The Independent, May 2023
- Yoshua Bengio: 50% · ABC News, Jul 2023
- Zvi Mowshowitz: 70% · The Cognitive Revolution, Sep 2025
- Daniel Kokotajlo: 70–80% · Interview, Apr 2025
- Dan Hendrycks: 80% · Fast Company, Jul 2023
- Andrew Critch: 85% · Doom Debates, Nov 2024
- Connor Leahy: 90% · Interview, Nov 2024
- Max Tegmark: 90% · Post on X, Apr 2025
- Eliezer Yudkowsky: 95% · Fast Company, Jul 2023
- Roman Yampolskiy: 99.9% · Business Insider, 2024
- Survey: Median of 2,778 AI researchers surveyed (Grace et al., 2024).
- Compiled from Wikipedia’s P(doom) article, which cites each original. Hinton has also said his own gut figure is over 50%; we plot the 10–20% he gives “all things considered”.
Where does the term p(doom) come from?
It started as informal slang among people who work on and argue about AI risk, as a quick way to ask “how worried are you?” It spread into interviews and headlines once prominent researchers and executives began giving their own figures in public.
Why do p(doom) estimates differ so much?
- “Doom” is not defined. Some people mean human extinction, others mean permanent loss of control, others something milder. Two people can give the same number and mean different things.
- The time frame varies. Ten years, this century, ever.
- It is a judgement, not a measurement. Estimates given in public run from near zero to near certain, and the same person’s figure can move with the news.
What can p(doom) not tell you?
It cannot be verified or recomputed, and it does not tell you what is happening today. A room of experts with very different figures can all be looking at the same facts.
How is SIREN different from p(doom)?
SIREN does not estimate the chance of anything. It counts how much is happening in AI, hourly, and reports it as a level from 5 (quietest) to 1 (loudest). Every reading ships with a receipt, and a free tool lets you re-check the published record in your own browser. The site’s line is: everyone has a p(doom), nobody has a receipt.
Right now: SIREN 4, ROUTINE, composite 56.6 of 100, observed 2026-10-15 05:21:57 UTC. AI activity index today · SIREN →
What questions do people actually ask?
What is p(doom)?
p(doom) is shorthand for "probability of doom": one person's estimate, as a percentage, that advanced AI leads to a catastrophic outcome for humanity. It is an opinion expressed as a number, with no agreed method behind it.
Where does the term p(doom) come from?
It started as informal slang among people who work on and argue about AI risk, as a quick way to ask “how worried are you?” It spread into interviews and headlines once prominent researchers and executives began giving their own figures in public.
Why do p(doom) estimates differ so much?
Because "doom" is not defined the same way by everyone, the time frame varies, and the figure is a judgement rather than a measurement.
What can p(doom) not tell you?
It cannot be verified or recomputed, and it does not tell you what is happening today. A room of experts with very different figures can all be looking at the same facts.
How is SIREN different from p(doom)?
SIREN does not estimate the chance of anything. It counts how much is happening in AI, hourly, and reports it as a level from 5 (quietest) to 1 (loudest). Everyone has a p(doom); nobody has a receipt.
Does SIREN publish a p(doom) or a doom probability?
No. SIREN measures activity tempo — how much is happening in AI this hour — not danger and not a probability of harm. Everyone has a p(doom); nobody has a receipt.
What else should you read about p(doom)?
Compare SIREN to DEFCON, the Doomsday Clock and p(doom); read the AI doomsday clock explainer; browse AI books that argue both sides; or open the history of AI risk.
What is Yann LeCun’s p(doom)?
Yann LeCun (Chief AI Scientist, Meta (at the time)) has put it at 0.01% (TechRadar, Apr 2024). People revise these figures; SIREN plots it beside 27 other named estimates.
What is Vitalik Buterin’s p(doom)?
Vitalik Buterin (Co-founder, Ethereum) has put it at 12% (Doom Debates, Aug 2025). People revise these figures; SIREN plots it beside 27 other named estimates.
What is Lina Khan’s p(doom)?
Lina Khan (Former FTC chair) has put it at 15% (New York Times, Dec 2023). People revise these figures; SIREN plots it beside 27 other named estimates.
What is Geoffrey Hinton’s p(doom)?
Geoffrey Hinton (Nobel laureate, “godfather of AI”) has put it at 10–20% (METR Q&A, Jun 2024). People revise these figures; SIREN plots it beside 27 other named estimates.
What is Dario Amodei’s p(doom)?
Dario Amodei (CEO, Anthropic) has put it at 10–25% (Axios, Sep 2025). People revise these figures; SIREN plots it beside 27 other named estimates.
What is Elon Musk’s p(doom)?
Elon Musk (CEO, xAI, Tesla, SpaceX) has put it at 10–30% (Business Insider, 2024). People revise these figures; SIREN plots it beside 27 other named estimates.
What is Yoshua Bengio’s p(doom)?
Yoshua Bengio (Turing Award winner, Mila) has put it at 50% (ABC News, Jul 2023). People revise these figures; SIREN plots it beside 27 other named estimates.
What is Max Tegmark’s p(doom)?
Max Tegmark (MIT physicist, Future of Life Institute (“over 90%”)) has put it at 90% (Post on X, Apr 2025). People revise these figures; SIREN plots it beside 27 other named estimates.
What is Eliezer Yudkowsky’s p(doom)?
Eliezer Yudkowsky (Founder, MIRI (“over 95%”)) has put it at 95% (Fast Company, Jul 2023). People revise these figures; SIREN plots it beside 27 other named estimates.
What is Roman Yampolskiy’s p(doom)?
Roman Yampolskiy (Computer scientist (99.9% or higher)) has put it at 99.9% (Business Insider, 2024). People revise these figures; SIREN plots it beside 27 other named estimates.
What do AI researchers think the chance of AI doom is?
In the largest survey, 2,778 AI researchers gave a median of 5% (mean 9%) for an outcome as bad as human extinction (Grace et al., 2024). Named public figures range from under 0.01% (Yann LeCun) to over 95% (Eliezer Yudkowsky).
Where do 28 named p(doom) estimates sit on one ladder?
Each bar is one person’s publicly stated p(doom), with where they said it. A range means they gave a range. Figures as compiled 2026-03; people revise them. SIREN does not publish a p(doom).
What is in the news about AI catastrophe, safety, and p(doom)?
Refreshes live from SIREN’s newsroom. Catastrophe-risk headlines when present; otherwise the newsroom says none landed in the last few days.
Which polls and surveys report p(doom)?
Named polls and surveys are listed under that heading on /p-doom.html, each with a figure and a source link.
What further reading covers every side?
Further reading on every side is listed under that heading on /p-doom.html, with a link into the library for books that argue both sides.
Search-shaped sibling: How does SIREN compare to DEFCON, the Doomsday Clock and p(doom)? · Is there an AI doomsday clock? · How the SIREN AI activity index is computed · methodology.
What else should you read about p(doom)?
How does SIREN compare to DEFCON, the Doomsday Clock and p(doom)? → · Is there an AI doomsday clock? → · What AI books argue both sides of the case? → · What is the history of AI risk? →