Skip to main content

Free public instrument from GAGE

The Doom Number

As of September 2026 this ledger holds 24 records across 3 jurisdictions: 11 verified at a primary source, 10 reported by a secondary source, 0 announced with no document yet, 1 searched and absent, and 2 open questions. By kind: stated probability 8, characterization 8, report finding 1, survey 2, signed statement 3, open question 2.

24
Records

3 jurisdictions, 18 primary sources

11
Verified at the source

10 more reported, primary not reached

1
Announced or absent

0 announced with no document, 1 searched and absent

2
Open questions

Posed, sourced, not answered

As of 16 September 2026 the GAGE Doom Number records 11 verified instruments, 10 reported at a secondary source, 0 announcements without a document, 1 absences and 2 open questions, across 3 jurisdictions. Counts are a floor, not a ceiling: an instrument the ledger has not found is not on it.

Why this ledger exists

A stated probability is a fact about the person who stated it

When a researcher says there is a ten percent chance of catastrophe, the record can verify one thing: that this person said this number, in this role, on this date, in this place. It cannot verify the number. The rows here are therefore verdicts about statements, fetched where they were published, and the kind on each row says whether it is a personal estimate, a characterization in words, a multi author report's finding, a survey median with public methodology, or a signed statement.

The numbers vary by two orders of magnitude between named experts, and the ledger does not average them, because the rows are not comparable: different timeframes, different definitions of catastrophe, different bases. What a reader can take from the page is who has said what, and how far each statement traveled from its source before it reached them.

Every row is fetched at its source where the source could be reached, and says so where it could not. The verdict language is the discipline: a reader learns five words once and never has to guess what a row claims.

  • Verified

    The document exists. The ledger fetched it at its publisher and quotes it.

  • Reported, primary not reached

    A reliable secondary source carries it, and the primary document could not be reached. Printed with this label, never as verified.

  • Announced, no document yet

    A body has said it will act. No document exists yet, so the row records the statement and nothing more.

  • Absent

    The ledger searched and found no instrument. The record says where it looked and when.

  • Open question

    No settled answer exists. The ledger poses the question, links the live debate, and does not answer it.

The grid

What the documents state, and where they are silent

For each dimension of a statement (the role at the time, where it was said, the timeframe named, the stated basis, any later revision), how many rows state it from the source and how many leave it open.

  • Role at the time

    11 stated, 0 silent, 1 open, 11 reported.

  • Where it was said

    11 stated, 1 silent, 10 reported.

  • Timeframe named

    11 stated, 1 silent, 1 open, 10 reported.

  • Stated basis

    11 stated, 1 silent, 3 open, 9 reported.

  • Later revision

    5 stated, 0 silent, 19 open.

Side by side

Every jurisdiction, counted by verdict and by kind

  • United States

    14 records

    Verified
    5
    Reported
    8
    Announced
    0
    Absent
    1
    Open question
    0

    6 stated probability, 7 characterization, 1 signed statement.

  • Canada

    3 records

    Verified
    1
    Reported
    2
    Announced
    0
    Absent
    0
    Open question
    0

    2 stated probability, 1 characterization.

  • Global

    7 records

    Verified
    5
    Reported
    0
    Announced
    0
    Absent
    0
    Open question
    2

    1 report finding, 2 survey, 2 signed statement, 2 open question.

Figures of record

Every number on this ledger, with who measured it and when

22 figures, each one printed in the unit its publisher used, beside the publisher and the date it was true. Nothing here is summed across sources, converted between units, or forecast.

  1. 31,810 signatories

    Signatories printed on the letter's page. Future of Life Institute: pause for at least six months the training of systems more powerful than GPT-4, March 2023

    Future of Life Institute, Pause Giant AI Experiments: An Open Letter, primary source, as of .

  2. 1,386 signatories

    Signatories printed on the statement's page. Pacing the Frontier statement: 1,386 industry signatories ask for tools to pace automated AI development, July 2026

    Pacing the Frontier, the statement and signatories, primary source, as of .

  3. 10 percent

    Stated lower bound of the probability that AI kills all humans within the next decade. Evan Hubinger: greater than 10 percent chance AI kills all humans within a decade, September 2026

    TechCrunch, Gambling with our lives: Anthropic researcher quits, warns against self improving AI, secondary source, as of .

  4. 70 percent

    Stated probability that AI development goes horribly wrong, including AI takeover, no timeframe stated. Daniel Kokotajlo: something like 70 percent that this goes horribly wrong, July 2026

    The Diary of a CEO, OpenAI whistleblower finally speaks: AI has a 70 percent chance of going horribly wrong (episode page), primary source, as of .

  5. 20 percent

    Reported upper bound of the risk that AI eventually takes control from humans, no timeframe stated. Geoffrey Hinton: 10 to 20 percent risk that AI eventually takes control from humans, April 2025

    CBS News, Godfather of AI Geoffrey Hinton warns AI could take control from humans, secondary source, as of .

  6. 10 percent

    Reported lower bound of the risk that AI eventually takes control from humans, no timeframe stated. Geoffrey Hinton: 10 to 20 percent risk that AI eventually takes control from humans, April 2025

    CBS News, Godfather of AI Geoffrey Hinton warns AI could take control from humans, secondary source, as of .

  7. 20 percent

    Stated upper bound of the chance that it goes bad, no timeframe stated. Elon Musk: some chance, 10 to 20 percent, that it goes bad, October 2024

    Fortune, Elon Musk says AI could go bad and be an existential threat, as xAI raises money, secondary source, as of .

  8. 10 percent

    Stated lower bound of the chance that it goes bad, no timeframe stated. Elon Musk: some chance, 10 to 20 percent, that it goes bad, October 2024

    Fortune, Elon Musk says AI could go bad and be an existential threat, as xAI raises money, secondary source, as of .

  9. 2,778 responses

    Responses received. Grace et al. 2023 survey of 2,778 AI authors: median 5 percent for extremely bad outcomes, October 2023

    Grace et al., Thousands of AI Authors on the Future of AI (arXiv 2401.02843), primary source, as of .

  10. 51 percent of respondents

    Upper share, across question framings, giving at least 10 percent to outcomes as bad as extinction. Grace et al. 2023 survey of 2,778 AI authors: median 5 percent for extremely bad outcomes, October 2023

    Grace et al., Thousands of AI Authors on the Future of AI (arXiv 2401.02843), primary source, as of .

  11. 38 percent of respondents

    Share of participants giving at least a 10 percent chance of extremely bad outcomes. Grace et al. 2023 survey of 2,778 AI authors: median 5 percent for extremely bad outcomes, October 2023

    Grace et al., Thousands of AI Authors on the Future of AI (arXiv 2401.02843), primary source, as of .

  12. 5 percent

    Median probability of extremely bad outcomes such as human extinction. Grace et al. 2023 survey of 2,778 AI authors: median 5 percent for extremely bad outcomes, October 2023

    Grace et al., Thousands of AI Authors on the Future of AI (arXiv 2401.02843), primary source, as of .

  13. 25 percent

    Stated upper bound of the probability that things go badly, no timeframe stated. Dario Amodei: 10 to 25 percent chance that things go badly, October 2023

    The Logan Bartlett Show, Dario Amodei's AI Predictions Through 2030 (episode 82 page), primary source, as of .

  14. 10 percent

    Stated lower bound of the probability that things go badly, no timeframe stated. Dario Amodei: 10 to 25 percent chance that things go badly, October 2023

    The Logan Bartlett Show, Dario Amodei's AI Predictions Through 2030 (episode 82 page), primary source, as of .

  15. 20 percent

    Stated probability that it turns out catastrophic, no timeframe stated for the outcome. Yoshua Bengio: around 20 percent probability that it turns out catastrophic, July 2023

    ABC News (Australia), What's your p(doom)? How AI could be learning a deceptive trick, secondary source, as of .

  16. 50 percent

    Stated chance of doom shortly after human level AI systems exist, given as 50 50. Paul Christiano: 10 to 20 percent chance of AI takeover with many or most humans dead, April 2023

    Fortune, OpenAI's former top safety researcher says there is a 10 to 20 percent chance the tech will take over, secondary source, as of .

  17. 20 percent

    Stated upper bound of the chance of AI takeover with many or most humans dead, no timeframe stated. Paul Christiano: 10 to 20 percent chance of AI takeover with many or most humans dead, April 2023

    Fortune, OpenAI's former top safety researcher says there is a 10 to 20 percent chance the tech will take over, secondary source, as of .

  18. 10 percent

    Stated lower bound of the chance of AI takeover with many or most humans dead, no timeframe stated. Paul Christiano: 10 to 20 percent chance of AI takeover with many or most humans dead, April 2023

    Fortune, OpenAI's former top safety researcher says there is a 10 to 20 percent chance the tech will take over, secondary source, as of .

  19. 738 responses

    Responses received. AI Impacts 2022 expert survey: median 5 percent for an extremely bad outcome such as human extinction, August 2022

    AI Impacts, 2022 Expert Survey on Progress in AI, primary source, as of .

  20. 10 percent

    Median probability that human failure to control AI causes extinction. AI Impacts 2022 expert survey: median 5 percent for an extremely bad outcome such as human extinction, August 2022

    AI Impacts, 2022 Expert Survey on Progress in AI, primary source, as of .

  21. 48 percent of respondents

    Share of respondents giving at least a 10 percent chance of an extremely bad outcome. AI Impacts 2022 expert survey: median 5 percent for an extremely bad outcome such as human extinction, August 2022

    AI Impacts, 2022 Expert Survey on Progress in AI, primary source, as of .

  22. 5 percent

    Median probability that the long run effect of advanced AI is extremely bad, for example human extinction. AI Impacts 2022 expert survey: median 5 percent for an extremely bad outcome such as human extinction, August 2022

    AI Impacts, 2022 Expert Survey on Progress in AI, primary source, as of .

The ledger

Every record, newest first

24 records. Each row opens a page carrying the answer, the verdict and what it means, the key facts, the figures with their sources, what it changes for a reader deciding how much weight to give a number, and the sources it was verified against.

RecordVerdictWhereKindVerified
Jacob Coxon: a 10 percent extinction estimate attributed to him by AP, no such statement found, September 2026ODDS-2026-0024AbsentUnited StatesStated probability
Open question: when a frontier lab researcher states a personal probability, what does it tell you about the lab?ODDS-2026-0023Open questionGlobalOpen question
Open question: are two named people's percentages comparable when neither states a timeframe or a definition of catastrophe?ODDS-2026-0022Open questionGlobalOpen question
Grace et al. 2023 survey of 2,778 AI authors: median 5 percent for extremely bad outcomes, October 2023ODDS-2026-0021VerifiedGlobalSurvey
AI Impacts 2022 expert survey: median 5 percent for an extremely bad outcome such as human extinction, August 2022ODDS-2026-0020VerifiedGlobalSurvey
Pacing the Frontier statement: 1,386 industry signatories ask for tools to pace automated AI development, July 2026ODDS-2026-0019VerifiedUnited StatesSigned statement
Future of Life Institute: pause for at least six months the training of systems more powerful than GPT-4, March 2023ODDS-2026-0018VerifiedGlobalSigned statement
Center for AI Safety Statement on AI Risk: extinction risk a global priority alongside pandemics and nuclear war, May 2023ODDS-2026-0017VerifiedGlobalSigned statement
Paul Christiano: 10 to 20 percent chance of AI takeover with many or most humans dead, April 2023ODDS-2026-0016Reported, primary not reachedUnited StatesStated probability
Daniel Kokotajlo: something like 70 percent that this goes horribly wrong, July 2026ODDS-2026-0015VerifiedUnited StatesStated probability
Jack Clark: we have seen warning shots and that gives us a window to act, September 2026ODDS-2026-0014Reported, primary not reachedUnited StatesCharacterization
Walter Isaacson: this is the first thing that just totally scares me, July 2026ODDS-2026-0013Reported, primary not reachedUnited StatesCharacterization
Elon Musk: some chance, 10 to 20 percent, that it goes bad, October 2024ODDS-2026-0012Reported, primary not reachedUnited StatesStated probability
Sam Altman: the pacing of AI development should be slower than it otherwise could be, September 2026ODDS-2026-0011Reported, primary not reachedUnited StatesCharacterization
Eliezer Yudkowsky: the most likely result of building superhumanly smart AI is that everyone dies, March 2023ODDS-2026-0010VerifiedUnited StatesCharacterization
Yoshua Bengio: even a 1 percent chance of something going really bad is not acceptable, April 2026ODDS-2026-0009VerifiedCanadaCharacterization
Yoshua Bengio: around 20 percent probability that it turns out catastrophic, July 2023ODDS-2026-0008Reported, primary not reachedCanadaStated probability
Geoffrey Hinton: 10 to 20 percent risk that AI eventually takes control from humans, April 2025ODDS-2026-0007Reported, primary not reachedCanadaStated probability
Dario Amodei: 10 to 25 percent chance that things go badly, October 2023ODDS-2026-0006VerifiedUnited StatesStated probability
Dario Amodei: we must slow the pace at which we improve AI capabilities, September 2026ODDS-2026-0005VerifiedUnited StatesCharacterization
Adam Khoja: current models are basically nation state level hackers, September 2026ODDS-2026-0004Reported, primary not reachedUnited StatesCharacterization
International AI Safety Report 2026: likelihood, nature and timing of loss of control unusually ambiguous, February 2026ODDS-2026-0003VerifiedGlobalReport finding
Jacob Coxon: labs are racing to self improving superintelligence and gambling with our lives, September 2026ODDS-2026-0002Reported, primary not reachedUnited StatesCharacterization
Evan Hubinger: greater than 10 percent chance AI kills all humans within a decade, September 2026ODDS-2026-0001Reported, primary not reachedUnited StatesStated probability

Answers

What people ask the Doom Number

What is p(doom)?

An individual's stated probability that advanced AI causes an existential catastrophe. It is a personal estimate, not a measurement: nobody has observed the event, so the number reports the speaker's belief and its stated basis, and values vary by orders of magnitude between named experts. The rows on this page record who stated what, when and where.

Who has stated a probability of AI catastrophe?

8 named persons on the ledger as of September 2026: Jacob Coxon: a 10 percent extinction estimate attributed to him by AP, no such statement found, September 2026; Paul Christiano: 10 to 20 percent chance of AI takeover with many or most humans dead, April 2023; Daniel Kokotajlo: something like 70 percent that this goes horribly wrong, July 2026; Elon Musk: some chance, 10 to 20 percent, that it goes bad, October 2024; Yoshua Bengio: around 20 percent probability that it turns out catastrophic, July 2023; Geoffrey Hinton: 10 to 20 percent risk that AI eventually takes control from humans, April 2025; and more. Each row carries the role at the time, where it was said, the timeframe and the stated basis.

Why does the page not average the numbers?

Because the rows are not comparable. One speaker names a decade and another names a century; one means extinction and another means a loss of control; one states a basis and another states none. An average of those would be a number nobody said, so the ledger prints none.

What do the reports and surveys say?

1 report finding and 2 surveys with public methodology are on the ledger, labeled apart from individual estimates. 2 rows record a later revision by the same person.

What number did a named person put on catastrophe, and when?

8 records of kind stated probability are on the Doom Number as of September 2026: Jacob Coxon: a 10 percent extinction estimate attributed to him by AP, no such statement found, September 2026; Paul Christiano: 10 to 20 percent chance of AI takeover with many or most humans dead, April 2023; Daniel Kokotajlo: something like 70 percent that this goes horribly wrong, July 2026; Elon Musk: some chance, 10 to 20 percent, that it goes bad, October 2024; Yoshua Bengio: around 20 percent probability that it turns out catastrophic, July 2023; Geoffrey Hinton: 10 to 20 percent risk that AI eventually takes control from humans, April 2025; and 2 more. Each has its own page with the verdict, the facts, the sources it was checked against and the date.

What did a named person say in words, without a number?

8 records of kind characterization are on the Doom Number as of September 2026: Jack Clark: we have seen warning shots and that gives us a window to act, September 2026; Walter Isaacson: this is the first thing that just totally scares me, July 2026; Sam Altman: the pacing of AI development should be slower than it otherwise could be, September 2026; Eliezer Yudkowsky: the most likely result of building superhumanly smart AI is that everyone dies, March 2023; Yoshua Bengio: even a 1 percent chance of something going really bad is not acceptable, April 2026; Dario Amodei: we must slow the pace at which we improve AI capabilities, September 2026; and 2 more. Each has its own page with the verdict, the facts, the sources it was checked against and the date.

What did a multi author report conclude about likelihood or timing?

1 record of kind report finding is on the Doom Number as of September 2026: International AI Safety Report 2026: likelihood, nature and timing of loss of control unusually ambiguous, February 2026. Each has its own page with the verdict, the facts, the sources it was checked against and the date.

What did a survey with public methodology find, and of whom?

2 records of kind survey are on the Doom Number as of September 2026: Grace et al. 2023 survey of 2,778 AI authors: median 5 percent for extremely bad outcomes, October 2023; AI Impacts 2022 expert survey: median 5 percent for an extremely bad outcome such as human extinction, August 2022. Each has its own page with the verdict, the facts, the sources it was checked against and the date.

What did a group of named people sign, and when?

3 records of kind signed statement are on the Doom Number as of September 2026: Pacing the Frontier statement: 1,386 industry signatories ask for tools to pace automated AI development, July 2026; Future of Life Institute: pause for at least six months the training of systems more powerful than GPT-4, March 2023; Center for AI Safety Statement on AI Risk: extinction risk a global priority alongside pandemics and nuclear war, May 2023. Each has its own page with the verdict, the facts, the sources it was checked against and the date.

What has nobody settled about these numbers?

2 records of kind open question are on the Doom Number as of September 2026: Open question: when a frontier lab researcher states a personal probability, what does it tell you about the lab?; Open question: are two named people's percentages comparable when neither states a timeframe or a definition of catastrophe?. Each has its own page with the verdict, the facts, the sources it was checked against and the date.

What do the verdicts mean?

Verified: the document exists and the ledger fetched it at its publisher. Reported: a reliable secondary source carries it and the primary could not be reached. Announced: a body said it will act and no document exists. Absent: the ledger searched and found nothing, and the search is written into the record. Open: a question nobody has settled, posed and not answered.

How current is the Doom Number?

Every record carries the date it was last verified; the ledger as a whole was last verified 16 September 2026 and holds 24 records with 18 primary sources. A change moves the record's own date and appears on the changelog, so a reader who cited a row can see whether it moved.

Every surface

Cut the ledger the way you need it

By jurisdiction

By kind

Every record page

Take the data

The whole dataset, free, in two formats

Licensed CC BY 4.0. Use it in an article, a paper, a slide or a product. The only condition is attribution, and the citation page gives you the line to paste.

How the ledger is built, what the verdicts mean, and what the gate refuses: the method page. Every change, dated: the changelog. The kinds on the shelf: stated probability, characterization, report finding, survey, signed statement, open question. Something missing or wrong is a bug, and we want to hear about it. A number being repeated without its source is graded on Settled or Not. What the people quoted here have asked governments to do is on the Slowdown Docket. The whole lane, with its terms, is at the Frontier Risk Lane.

Cite this page

Free to reuse under CC BY 4.0, with attribution.

In a sentence
According to the GAGE Doom Number (as of 16 September 2026), doom number.
APA
GAGE (Global Academy of Generative-AI Education). (2026). Doom Number. Doom Number. Retrieved 16 September 2026, from https://www.gage.academy/tools/doom-number
MLA
"Doom Number." Doom Number, GAGE (Global Academy of Generative-AI Education), 16 September 2026, https://www.gage.academy/tools/doom-number.
Chicago
GAGE (Global Academy of Generative-AI Education). "Doom Number." Doom Number. Last modified 16 September 2026. https://www.gage.academy/tools/doom-number.
Permalink
https://www.gage.academy/tools/doom-number

Last updated . Every record re verified . The ledger is checked weekly, every Monday, and the same day a named person states a new figure in public.

31,810 signatories

GAGE briefings tell you which AI regulation deadlines are coming, what they actually require of you, and when a program opens.