{ "meta": { "kit": "ledger-kit", "buildDate": "2026-09-16", "lastVerified": "2026-09-16", "license": "CC BY 4.0", "licenseUrl": "https://creativecommons.org/licenses/by/4.0/", "cadence": "weekly, every Monday, and the same day a named person states a new figure in public", "note": "Every record is a verdict about a statement. Verified means the statement was fetched where it was published (the interview transcript, the post, the report, the signed statement) on the date given and quoted. Reported means a reliable secondary source carries it and the primary could not be reached, and such a record carries no primary source. Absent means the ledger searched for a statement that is being attributed to someone and found none, and the search is written into the record. Open means a question nobody has settled. A row is a stated probability, a characterization in words, a report finding, a survey with public methodology, or a signed statement, one per statement, by a named person speaking in a public role, dated. The ledger computes no average: averaging the rows would imply they are comparable, and they are not. It makes no prediction. It records no anonymous poll unless the methodology is public. A number that later proves wrong stays, with its date, because that is the point of the page. Every figure carries the source that printed it and the date it was true, and no figure is estimated, summed across sources or converted.", "vocab": { "idPrefix": "ODDS", "root": "/tools/doom-number", "title": "The Doom Number", "short": "Doom Number", "claim": "What named people have actually said about the probability of catastrophe from AI, one row per statement, dated and sourced, with what the number means and why it is not a measurement. No average is computed.", "changesHeading": "For a reader deciding how much weight to give a number", "kinds": [ { "key": "probability", "label": "Stated probability", "question": "What number did a named person put on catastrophe, and when?" }, { "key": "characterization", "label": "Characterization", "question": "What did a named person say in words, without a number?" }, { "key": "report", "label": "Report finding", "question": "What did a multi author report conclude about likelihood or timing?" }, { "key": "survey", "label": "Survey", "question": "What did a survey with public methodology find, and of whom?" }, { "key": "statement", "label": "Signed statement", "question": "What did a group of named people sign, and when?" }, { "key": "question", "label": "Open question", "question": "What has nobody settled about these numbers?" } ], "jurisdictions": [ { "code": "US", "name": "United States" }, { "code": "UK", "name": "United Kingdom" }, { "code": "CA", "name": "Canada" }, { "code": "GLOBAL", "name": "Global" } ], "facets": [ { "key": "role", "label": "Role at the time" }, { "key": "context", "label": "Where it was said" }, { "key": "timeframe", "label": "Timeframe named" }, { "key": "basis", "label": "Stated basis" }, { "key": "later", "label": "Later revision" } ] }, "source": "https://www.gage.academy/tools/doom-number", "attribution": "GAGE (Global Academy of Generative-AI Education), Doom Number", "documentation": "https://www.gage.academy/tools/doom-number/data", "methodology": "https://www.gage.academy/tools/doom-number/method" }, "records": [ { "id": "ODDS-2026-0001", "slug": "evan-hubinger-greater-than-10-percent-within-decade-september-2026", "title": "Evan Hubinger: greater than 10 percent chance AI kills all humans within a decade, September 2026", "kind": "probability", "verdict": "reported", "jurisdiction": "US", "answer": "Evan Hubinger, who leads alignment science work at Anthropic, replied on X to Jacob Coxon's resignation post of 8 September 2026 and wrote that he personally thinks the chance AI kills all humans is greater than 10 percent within the next decade. TechCrunch and Time quoted the post on 9 September. The figure is one researcher's personal estimate, stated as a lower bound, not a measurement and not a company position.", "key_facts": [ "TechCrunch and Time quote the post: Hubinger wrote that Anthropic staff earnestly believe AI could kill all humans and that he personally puts it above 10 percent within the next decade.", "The same post says Anthropic is trying its best but does not yet have a plan to solve alignment for superintelligence and is not clearly on track to one.", "Time describes Hubinger as head of alignment stress testing at Anthropic; other reports call him alignment science lead.", "The post at x.com/EvanHub/status/2097497037956891126 returned 451 and 403 to the ledger's fetches on 16 September 2026, so the statement rests on the outlets that quoted it.", "The figure is a personal estimate with a stated timeframe and a stated outcome; the post names no method behind the number." ], "figures": [ { "label": "Stated lower bound of the probability that AI kills all humans within the next decade", "value": 10, "unit": "percent", "as_of": "2026-09-09", "source": 0 } ], "facets": [ { "key": "role", "value": "Alignment science lead, Anthropic (Time: head of alignment stress testing)", "verdict": "reported", "source": 1 }, { "key": "context", "value": "Post on X replying to Jacob Coxon's resignation post, quoted by TechCrunch and Time on 9 September 2026", "verdict": "reported", "source": 0 }, { "key": "timeframe", "value": "Within the next decade", "verdict": "reported", "source": 0 }, { "key": "basis", "value": "None stated; the post gives the number as a personal belief", "verdict": "reported", "source": 0 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A lower bound from one named researcher tells you what that person believes and how they choose to say it in public; it does not tell you how the number was reached, what would move it, or whether colleagues share it. The stated timeframe and outcome make this one of the more specific statements on the ledger, which is what lets a reader compare it with another statement at all. Weigh it as testimony about a belief, not as a measured rate.", "sources": [ { "name": "TechCrunch, Gambling with our lives: Anthropic researcher quits, warns against self improving AI", "url": "https://techcrunch.com/2026/09/09/gambling-with-our-lives-anthropic-researcher-quits-warns-against-self-improving-ai/", "type": "secondary", "date": "2026-09-09" }, { "name": "Time, He helped build powerful AI at OpenAI and Anthropic. Now he's afraid it could kill us", "url": "https://time.com/article/2026/09/09/ai-anthropic-openai-jacob-coxon/", "type": "secondary", "date": "2026-09-09" }, { "name": "LessWrong, Jacob Coxon warns of human extinction and triggers a debate", "url": "https://www.lesswrong.com/posts/5MB7KENgEAW6Q4JtJ/jacob-coxon-warns-of-human-extinction-and-triggers-a", "type": "secondary", "date": "2026-09-10" } ], "related_ids": [ "ODDS-2026-0002", "ODDS-2026-0024", "ODDS-2026-0005" ], "tags": [ "Evan Hubinger", "Anthropic", "Coxon resignation", "personal estimate", "within a decade" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0002", "slug": "jacob-coxon-racing-to-self-improving-superintelligence-september-2026", "title": "Jacob Coxon: labs are racing to self improving superintelligence and gambling with our lives, September 2026", "kind": "characterization", "verdict": "reported", "jurisdiction": "US", "answer": "Jacob Coxon, a pretraining researcher who spent about three years at OpenAI and then Anthropic, resigned from Anthropic on 8 September 2026 and posted on X that neither company is acting responsibly and that they are racing straight to self improving superintelligence and gambling with our lives. Time and TechCrunch quote the post. It is a characterization in words; the reached reports carry no number from him.", "key_facts": [ "Time dates the resignation to 8 September 2026 and describes roughly three years of pretraining research at OpenAI and then Anthropic, which he joined earlier in 2026.", "The post, as quoted by TechCrunch and Time: neither company is acting responsibly; they are racing straight to self improving superintelligence and gambling with our lives.", "Time quotes him in interview: it is obvious that things are speeding up, and they are not under control.", "Neither Time nor TechCrunch prints a probability from Coxon; the greater than 10 percent figure in the same news cycle is Evan Hubinger's (ODDS-2026-0001).", "The post at x.com/hilbertspaess/status/2097476196791709843 returned 451 to the ledger's fetch on 16 September 2026." ], "figures": [], "facets": [ { "key": "role", "value": "Pretraining researcher, formerly OpenAI, resigned from Anthropic on 8 September 2026", "verdict": "reported", "source": 1 }, { "key": "context", "value": "Post on X on 8 September 2026 and an interview quoted by Time on 9 September 2026", "verdict": "reported", "source": 1 }, { "key": "timeframe", "value": "None stated in the post; the interview quotes speak of things speeding up now", "verdict": "reported", "source": 1 }, { "key": "basis", "value": "His own account of pretraining work at both companies", "verdict": "reported", "source": 0 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A characterization by a departing employee tells you what one insider concluded and chose to say on the way out; it carries no number, so it cannot be compared with any stated probability, and it should not be given one by a reader or a headline. What it can tell you is that a person with direct exposure to the training work reached this view in September 2026, which is a fact about the field's own disagreement rather than about the odds themselves.", "sources": [ { "name": "TechCrunch, Gambling with our lives: Anthropic researcher quits, warns against self improving AI", "url": "https://techcrunch.com/2026/09/09/gambling-with-our-lives-anthropic-researcher-quits-warns-against-self-improving-ai/", "type": "secondary", "date": "2026-09-09" }, { "name": "Time, He helped build powerful AI at OpenAI and Anthropic. Now he's afraid it could kill us", "url": "https://time.com/article/2026/09/09/ai-anthropic-openai-jacob-coxon/", "type": "secondary", "date": "2026-09-09" }, { "name": "LessWrong, Jacob Coxon warns of human extinction and triggers a debate", "url": "https://www.lesswrong.com/posts/5MB7KENgEAW6Q4JtJ/jacob-coxon-warns-of-human-extinction-and-triggers-a", "type": "secondary", "date": "2026-09-10" } ], "related_ids": [ "ODDS-2026-0001", "ODDS-2026-0024", "ODDS-2026-0005" ], "tags": [ "Jacob Coxon", "Coxon resignation", "Anthropic", "OpenAI", "characterization" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0003", "slug": "international-ai-safety-report-2026-loss-of-control-unusually-ambiguous", "title": "International AI Safety Report 2026: likelihood, nature and timing of loss of control unusually ambiguous, February 2026", "kind": "report", "verdict": "verified", "jurisdiction": "GLOBAL", "answer": "The International AI Safety Report 2026, published 3 February 2026 under the chair Yoshua Bengio with more than 100 contributing experts and nominees from over 30 countries, writes on page 76 that a key challenge for policymakers is preparing for a risk whose likelihood, nature and timing remains unusually ambiguous. It records that expert opinion on loss of control varies greatly. The report states no probability; the finding is a characterization of disagreement.", "key_facts": [ "Page 76 of the report reads that a key challenge for policymakers is preparing for a risk whose likelihood, nature, and timing remains unusually ambiguous; the ledger read the sentence in the PDF at arXiv 2602.21012 and at the report's own site.", "The same page defines loss of control scenarios as scenarios in which one or more general purpose AI systems operate outside of anyone's control, and regaining control is either extremely costly or impossible.", "Page 76 states that expert opinion on the likelihood of loss of control varies greatly, and that some experts consider such scenarios implausible.", "Page 77 records that experts disagree about the likelihood and potential severity, and that some believe outcomes as extreme as the extinction of humanity are plausible while others think such outcomes implausible.", "The report's site gives the publication date as 3 February 2026, the chair as Yoshua Bengio, over 100 AI experts, and an advisory panel with nominees from more than 30 countries and international organisations." ], "figures": [], "facets": [ { "key": "role", "value": "Multi author scientific report chaired by Yoshua Bengio, commissioned through the international AI safety summit process", "verdict": "verified", "source": 1 }, { "key": "context", "value": "International AI Safety Report 2026, page 76, published 3 February 2026", "verdict": "verified", "source": 0 }, { "key": "timeframe", "value": "None stated; the report describes timing itself as ambiguous", "verdict": "verified", "source": 0 }, { "key": "basis", "value": "A synthesis of the published literature, with numbered references on pages 76 and 77", "verdict": "verified", "source": 0 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A report that says the likelihood is unusually ambiguous is telling you that the evidence does not support a number, which is a different claim from any single person's estimate on this ledger. When a headline pairs this report with a percentage, the two are not agreeing: the report is declining to give one. Read it as the most carefully reviewed statement here of how much is unsettled, and as a reason to treat every individual figure as a belief.", "sources": [ { "name": "International AI Safety Report 2026 (arXiv 2602.21012), page 76", "url": "https://arxiv.org/abs/2602.21012", "type": "primary", "date": "2026-02-24" }, { "name": "International AI Safety Report, publication page for the 2026 report", "url": "https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026", "type": "primary", "date": "2026-02-03" }, { "name": "International AI Safety Report 2026, the PDF at the report's own site", "url": "https://internationalaisafetyreport.org/sites/default/files/2026-02/international-ai-safety-report-2026_1.pdf", "type": "primary", "date": "2026-02-03" }, { "name": "PBS NewsHour (AP), What to know about recent dire AI predictions and calls for safeguards", "url": "https://www.pbs.org/newshour/science/what-to-know-about-recent-dire-ai-predictions-and-calls-for-safeguards", "type": "secondary", "date": "2026-09-14" } ], "related_ids": [ "ODDS-2026-0009", "ODDS-2026-0021", "ODDS-2026-0022" ], "tags": [ "International AI Safety Report", "Yoshua Bengio", "loss of control", "report finding", "expert disagreement" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0004", "slug": "adam-khoja-models-basically-nation-state-level-hackers-september-2026", "title": "Adam Khoja: current models are basically nation state level hackers, September 2026", "kind": "characterization", "verdict": "reported", "jurisdiction": "US", "answer": "Adam Khoja, a researcher at the Center for AI Safety in San Francisco, is quoted in a Las Vegas Sun piece of 11 September 2026 saying that at this point the models are basically nation state level hackers. The remark describes present capability in words and states no probability of catastrophe. The Sun page refused the ledger's fetches, so the quote rests on that outlet's own page as indexed.", "key_facts": [ "The Las Vegas Sun piece of 11 September 2026 quotes Khoja: at this point, the models are basically nation state level hackers.", "The Sun identifies him as a researcher at the Center for AI Safety in San Francisco; the Center's own pages list him as a special projects manager and researcher.", "The statement is about the capability of current models, not about the probability or timing of any outcome.", "The Sun's page returned 403 to the ledger's fetches on 16 September 2026; no post by Khoja carrying the sentence was found on his own site or on X." ], "figures": [], "facets": [ { "key": "role", "value": "Researcher, Center for AI Safety", "verdict": "reported", "source": 0 }, { "key": "context", "value": "Quoted in the Las Vegas Sun, 11 September 2026", "verdict": "reported", "source": 0 }, { "key": "timeframe", "value": "None stated; the remark is about the present", "verdict": "reported", "source": 0 }, { "key": "basis", "value": "None stated in the quoted passage", "verdict": "reported", "source": 0 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A capability characterization tells you what one researcher thinks a system can do today; it says nothing about how likely any outcome is, and a reader should not convert it into odds. Its use is narrower and more checkable than a probability: capability claims can be tested against published evaluations, which is where a reader who wants to weigh it should look next.", "sources": [ { "name": "Las Vegas Sun, What we learned about AI last week should terrify all of us", "url": "https://lasvegassun.com/news/2026/sep/11/what-we-learned-about-ai-last-week-should-terrify/", "type": "secondary", "date": "2026-09-11" } ], "related_ids": [ "ODDS-2026-0017", "ODDS-2026-0014" ], "tags": [ "Adam Khoja", "Center for AI Safety", "capability", "characterization", "cyber" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0005", "slug": "dario-amodei-we-must-pace-the-frontier-september-2026", "title": "Dario Amodei: we must slow the pace at which we improve AI capabilities, September 2026", "kind": "characterization", "verdict": "verified", "jurisdiction": "US", "answer": "Dario Amodei, chief executive of Anthropic, published the essay We Must Pace the Frontier on his own site on 12 September 2026. It opens with the claim that we must slow the pace at which we improve the capabilities of AI models, and sets out a three part plan. It describes a swarm of agents that in 6 to 12 months could take over the internet. The essay states no probability of catastrophe.", "key_facts": [ "The essay on darioamodei.com states: we must slow the pace at which we improve the capabilities of AI models, progress will still seem fast, and we must make wise use of the time we gain.", "It names two triggers: recursive self improvement accelerating since summer 2026, and the OpenAI agent swarm incident at Hugging Face.", "It describes a swarm that in 6 to 12 months could be capable of taking over the entire internet with a persistent botnet; this is a capability timeframe, not a stated probability.", "The three parts: embedded third party evaluators with employee level access (Anthropic commits unilaterally), common standards among frontier companies in democracies, then coordination with other governments.", "AP reporting of 14 September 2026 says OpenAI's Sam Altman wrote that pacing should be slower than it otherwise could be (ODDS-2026-0011)." ], "figures": [], "facets": [ { "key": "role", "value": "Chief executive, Anthropic", "verdict": "verified", "source": 0 }, { "key": "context", "value": "Essay on darioamodei.com, 12 September 2026", "verdict": "verified", "source": 0 }, { "key": "timeframe", "value": "6 to 12 months for the agent swarm capability he describes; none for any catastrophe probability", "verdict": "verified", "source": 0 }, { "key": "basis", "value": "Anthropic's internal observations of self improvement and the Hugging Face incident, as the essay states them", "verdict": "verified", "source": 0 }, { "key": "later", "value": "His 2023 range of 10 to 25 percent is a separate row (ODDS-2026-0006); the essay restates no number", "verdict": "verified", "source": 0 } ], "what_it_changes": "A chief executive's essay tells you what the head of a frontier company is willing to commit to in public and what he wants others to do; it is a policy position with commercial consequences, not a probability, and the reader should notice that it gives no number even where it gives a timeframe. Its weight comes from who wrote it and what he committed his company to, which is checkable, rather than from any estimate of odds.", "sources": [ { "name": "Dario Amodei, We Must Pace the Frontier", "url": "https://darioamodei.com/post/we-must-pace-the-frontier", "type": "primary", "date": "2026-09-12" }, { "name": "ABC News (AP), New warnings about the risks of AI to humanity revive a long running debate", "url": "https://abcnews.com/US/wireStory/new-warnings-risks-ai-humanity-revive-long-running-136414576", "type": "secondary", "date": "2026-09-14" } ], "related_ids": [ "ODDS-2026-0006", "ODDS-2026-0011", "ODDS-2026-0019", "ODDS-2026-0001" ], "tags": [ "Dario Amodei", "Amodei essay", "Pacing the Frontier", "Anthropic", "characterization" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0006", "slug": "dario-amodei-10-to-25-percent-chance-things-go-badly-october-2023", "title": "Dario Amodei: 10 to 25 percent chance that things go badly, October 2023", "kind": "probability", "verdict": "verified", "jurisdiction": "US", "answer": "Dario Amodei, chief executive of Anthropic, gave a range on The Logan Bartlett Show, episode 82, released 6 October 2023. The show's own episode page, dated 9 October 2023, records that he gives a 75 to 90 percent chance that the technology is built and everything goes fine and spends much of his effort reducing the 10 to 25 percent chance that disaster could occur. The range is a personal estimate with no stated timeframe.", "key_facts": [ "The Logan Bartlett Show's own page for episode 82 records: Dario gives a 75 to 90 percent chance that crazy tech will be built and everything will go fine, and works on reducing the 10 to 25 percent chance that disaster could occur.", "The episode was released on 6 October 2023; the show's written page is dated 9 October 2023.", "The page is the show's own summary, not a verbatim transcript; the ledger did not reach a transcript of the spoken words.", "No timeframe accompanies the range on the show's page.", "The essay of 12 September 2026 (ODDS-2026-0005) restates no number." ], "figures": [ { "label": "Stated lower bound of the probability that things go badly, no timeframe stated", "value": 10, "unit": "percent", "as_of": "2023-10-09", "source": 0 }, { "label": "Stated upper bound of the probability that things go badly, no timeframe stated", "value": 25, "unit": "percent", "as_of": "2023-10-09", "source": 0 } ], "facets": [ { "key": "role", "value": "Chief executive, Anthropic", "verdict": "verified", "source": 0 }, { "key": "context", "value": "The Logan Bartlett Show, episode 82, released 6 October 2023; the show's page dated 9 October 2023", "verdict": "verified", "source": 0 }, { "key": "timeframe", "value": "None stated", "verdict": "verified", "source": 0 }, { "key": "basis", "value": "None stated on the show's page", "verdict": "verified", "source": 0 }, { "key": "later", "value": "Not restated in We Must Pace the Frontier, 12 September 2026 (ODDS-2026-0005); no later figure found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A range from a chief executive tells you how the person running one frontier company frames the stakes of his own product, which is unusual testimony and worth recording; it does not tell you how the number was formed, and a range this wide with no timeframe cannot be compared with a bounded statement such as ODDS-2026-0001. Read the two ends as the person's own uncertainty, and never take a midpoint, because the speaker did not.", "sources": [ { "name": "The Logan Bartlett Show, Dario Amodei's AI Predictions Through 2030 (episode 82 page)", "url": "https://theloganbartlettshow.substack.com/p/dario-amodeis-ai-predictions-through", "type": "primary", "date": "2023-10-09" } ], "related_ids": [ "ODDS-2026-0005", "ODDS-2026-0007", "ODDS-2026-0016" ], "tags": [ "Dario Amodei", "Anthropic", "range", "personal estimate", "podcast" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0007", "slug": "geoffrey-hinton-10-to-20-percent-ai-takes-control-april-2025", "title": "Geoffrey Hinton: 10 to 20 percent risk that AI eventually takes control from humans, April 2025", "kind": "probability", "verdict": "reported", "jurisdiction": "CA", "answer": "Geoffrey Hinton, emeritus professor at the University of Toronto and 2024 Nobel laureate in physics, is reported by CBS News on 26 April 2025 to estimate a 10 to 20 percent risk that artificial intelligence will eventually take control from humans. CBS paraphrases rather than quotes the figure and names no timeframe. Earlier reports of the same range on BBC Radio 4 in December 2024 were not reached at a primary.", "key_facts": [ "CBS News, 26 April 2025 (CBS Mornings, Analisa Novak and Brook Silva-Braga): the AI pioneer estimates a 10 to 20 percent risk that artificial intelligence will eventually take control from humans.", "The CBS piece paraphrases the range; it prints no verbatim sentence from Hinton containing the numbers and states no timeframe.", "Hinton is the first named signatory of the May 2023 Center for AI Safety statement (ODDS-2026-0017).", "Reports of the same range with a 30 year timeframe on BBC Radio 4's Today programme, 27 December 2024, exist in outlets the ledger could not fetch; the programme audio was not reached." ], "figures": [ { "label": "Reported lower bound of the risk that AI eventually takes control from humans, no timeframe stated", "value": 10, "unit": "percent", "as_of": "2025-04-26", "source": 0 }, { "label": "Reported upper bound of the risk that AI eventually takes control from humans, no timeframe stated", "value": 20, "unit": "percent", "as_of": "2025-04-26", "source": 0 } ], "facets": [ { "key": "role", "value": "Emeritus professor of computer science, University of Toronto; Nobel laureate in physics 2024", "verdict": "reported", "source": 0 }, { "key": "context", "value": "CBS Mornings interview reported by CBS News, 26 April 2025", "verdict": "reported", "source": 0 }, { "key": "timeframe", "value": "None stated in the CBS report", "verdict": "reported", "source": 0 }, { "key": "basis", "value": "None stated in the CBS report", "verdict": "reported", "source": 0 }, { "key": "later", "value": "The range is reported as unchanged across 2024 and 2025 interviews; no later figure found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A range repeated by the same person across several interviews tells you the belief is stable, which is more than a single remark tells you; it still rests on a paraphrase here, so the exact words, outcome and timeframe are the reporter's framing. A reader weighing it should separate the stability of the belief, which is documented, from the precision of the number, which is not.", "sources": [ { "name": "CBS News, Godfather of AI Geoffrey Hinton warns AI could take control from humans", "url": "https://www.cbsnews.com/news/godfather-of-ai-geoffrey-hinton-ai-warning/", "type": "secondary", "date": "2025-04-26" } ], "related_ids": [ "ODDS-2026-0017", "ODDS-2026-0012", "ODDS-2026-0008" ], "tags": [ "Geoffrey Hinton", "range", "takes control", "CBS", "Nobel laureate" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0008", "slug": "yoshua-bengio-around-20-percent-probability-catastrophic-july-2023", "title": "Yoshua Bengio: around 20 percent probability that it turns out catastrophic, July 2023", "kind": "probability", "verdict": "reported", "jurisdiction": "CA", "answer": "Yoshua Bengio, professor at Universite de Montreal and scientific director of Mila, told ABC Radio National's Background Briefing, reported 15 July 2023, that he got around 20 percent probability that it turns out catastrophic. He named the inputs: about 50 percent that AI reaches human level within a decade, more than 50 percent that it is turned against humanity at scale, about 50 percent that defence succeeds. The broadcast itself was not reached.", "key_facts": [ "ABC News (Australia), Background Briefing, 15 July 2023, quotes Bengio: I got around, like, 20 per cent probability that it turns out catastrophic.", "The report records his inputs: about 50 percent that AI reaches human level capabilities within a decade, a greater than 50 percent chance that AI or humans turn the technology against humanity at scale, and about 50 percent that humans defend successfully.", "This is the only reached statement in which Bengio names a figure of his own; his June 2023 FAQ and his 2026 interviews give thresholds and confidence intervals, not a probability of catastrophe.", "The ABC page is the outlet's written account of the broadcast; the audio was not fetched." ], "figures": [ { "label": "Stated probability that it turns out catastrophic, no timeframe stated for the outcome", "value": 20, "unit": "percent", "as_of": "2023-07-15", "source": 0 } ], "facets": [ { "key": "role", "value": "Professor, Universite de Montreal; scientific director, Mila", "verdict": "reported", "source": 0 }, { "key": "context", "value": "ABC Radio National Background Briefing, reported by ABC News on 15 July 2023", "verdict": "reported", "source": 0 }, { "key": "timeframe", "value": "A decade for the human level input; none for the catastrophe itself", "verdict": "reported", "source": 0 }, { "key": "basis", "value": "Three inputs he named, each about one half, multiplied in his own words", "verdict": "reported", "source": 0 }, { "key": "later", "value": "By April 2026 he states a threshold, not a probability: even a 1 percent chance is not acceptable to him (ODDS-2026-0009)", "verdict": "verified", "source": 1 } ], "what_it_changes": "This is the rare row where the speaker shows the arithmetic, and that is what makes it useful: a reader can see which of the three inputs they would set differently and how the result would move. It remains one person's judgment on each input. The fact that the same person later chose to speak in thresholds rather than a probability is itself information about how much weight the speaker puts on the number.", "sources": [ { "name": "ABC News (Australia), What's your p(doom)? How AI could be learning a deceptive trick", "url": "https://www.abc.net.au/news/2023-07-15/whats-your-pdoom-ai-researchers-worry-catastrophe/102591340", "type": "secondary", "date": "2023-07-15" }, { "name": "80,000 Hours podcast, Yoshua Bengio thinks he knows how to build safe superintelligence (transcript)", "url": "https://80000hours.org/podcast/episodes/yoshua-bengio-scientist-ai/", "type": "secondary", "date": "2026-04-16" } ], "related_ids": [ "ODDS-2026-0009", "ODDS-2026-0003", "ODDS-2026-0007" ], "tags": [ "Yoshua Bengio", "personal estimate", "stated inputs", "Mila", "podcast" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0009", "slug": "yoshua-bengio-even-a-1-percent-chance-not-acceptable-april-2026", "title": "Yoshua Bengio: even a 1 percent chance of something going really bad is not acceptable, April 2026", "kind": "characterization", "verdict": "verified", "jurisdiction": "CA", "answer": "Yoshua Bengio, chair of the International AI Safety Report and founder of LawZero, said on the 80,000 Hours podcast published 16 April 2026 that even a 1 percent chance of something going really, really bad is not acceptable to him. The transcript is on the podcast's own site. The sentence sets a threshold he would act on; it is not an estimate of how likely the outcome is, and the ledger records no figure for it.", "key_facts": [ "The 80,000 Hours transcript, published 16 April 2026, quotes Bengio: even a 1 percent chance of something going really, really bad is not acceptable to me.", "The transcript contains no percentage estimate by Bengio of the probability of catastrophe; the remark is a tolerance threshold.", "His June 2023 FAQ on his own site gives a 95 percent confidence interval of 5 to 20 years for superhuman intelligence and argues for prudence under uncertainty, again without a probability of catastrophe.", "He chaired the International AI Safety Report 2026, which describes the likelihood of loss of control as unusually ambiguous (ODDS-2026-0003)." ], "figures": [], "facets": [ { "key": "role", "value": "Chair, International AI Safety Report; professor, Universite de Montreal; founder, LawZero", "verdict": "verified", "source": 0 }, { "key": "context", "value": "80,000 Hours podcast transcript, 16 April 2026", "verdict": "verified", "source": 0 }, { "key": "timeframe", "value": "None stated", "verdict": "verified", "source": 0 }, { "key": "basis", "value": "The magnitude of the outcome, as he argues in his 2023 FAQ", "verdict": "verified", "source": 1 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A threshold statement tells you what probability a person would refuse to accept, not what probability they believe; the two are often confused in headlines, and a 1 percent threshold is not a 1 percent estimate. It is useful to a reader as a statement of decision rule, which can be argued with directly, and it explains why the same person declines to publish a single number elsewhere.", "sources": [ { "name": "80,000 Hours podcast, Yoshua Bengio thinks he knows how to build safe superintelligence (transcript)", "url": "https://80000hours.org/podcast/episodes/yoshua-bengio-scientist-ai/", "type": "primary", "date": "2026-04-16" }, { "name": "Yoshua Bengio, FAQ on Catastrophic AI Risks", "url": "https://yoshuabengio.org/2023/06/24/faq-on-catastrophic-ai-risks/", "type": "primary", "date": "2023-06-24" } ], "related_ids": [ "ODDS-2026-0008", "ODDS-2026-0003", "ODDS-2026-0017" ], "tags": [ "Yoshua Bengio", "threshold", "characterization", "International AI Safety Report", "LawZero" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0010", "slug": "eliezer-yudkowsky-most-likely-result-everyone-on-earth-dies-march-2023", "title": "Eliezer Yudkowsky: the most likely result of building superhumanly smart AI is that everyone dies, March 2023", "kind": "characterization", "verdict": "verified", "jurisdiction": "US", "answer": "Eliezer Yudkowsky, of the Machine Intelligence Research Institute, wrote in Time on 29 March 2023 that the most likely result of building a superhumanly smart AI, under anything remotely like the current circumstances, is that literally everyone on Earth will die. He gave no percentage, adding that he means the obvious thing that would happen rather than a remote chance. The ledger records it as a characterization with no figure.", "key_facts": [ "Time, 29 March 2023, by Eliezer Yudkowsky, titled Pausing AI Developments Isn't Enough. We Need to Shut it All Down.", "The essay states that the most likely result of building a superhumanly smart AI, under anything remotely like the current circumstances, is that literally everyone on Earth will die.", "It gives no percentage, and frames the claim as the obvious thing that would happen, not as a remote chance.", "The essay was written in response to the Future of Life Institute pause letter of 22 March 2023 (ODDS-2026-0018), which he did not sign, arguing it asked for too little." ], "figures": [], "facets": [ { "key": "role", "value": "Co founder and research lead, Machine Intelligence Research Institute", "verdict": "verified", "source": 0 }, { "key": "context", "value": "Essay in Time, 29 March 2023", "verdict": "verified", "source": 0 }, { "key": "timeframe", "value": "None stated; conditional on building superhumanly smart AI under current circumstances", "verdict": "verified", "source": 0 }, { "key": "basis", "value": "His argument in the essay that alignment is unsolved and cannot be solved on the current timeline", "verdict": "verified", "source": 0 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A most likely claim without a number is a stronger statement than any percentage on this ledger and also a less comparable one, because it names no probability and no date, only a conditional. A reader weighing it should notice that the author deliberately refuses the framing of odds; the statement can be engaged on its argument, which is published, but it cannot be placed on a scale next to a survey median.", "sources": [ { "name": "Time, Eliezer Yudkowsky, Pausing AI Developments Isn't Enough. We Need to Shut it All Down", "url": "https://time.com/6266923/ai-eliezer-yudkowsky-open-letter-not-enough/", "type": "primary", "date": "2023-03-29" } ], "related_ids": [ "ODDS-2026-0018", "ODDS-2026-0015", "ODDS-2026-0022" ], "tags": [ "Eliezer Yudkowsky", "MIRI", "characterization", "Time essay", "no number" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0011", "slug": "sam-altman-pacing-should-be-slower-than-it-otherwise-could-be-september-2026", "title": "Sam Altman: the pacing of AI development should be slower than it otherwise could be, September 2026", "kind": "characterization", "verdict": "reported", "jurisdiction": "US", "answer": "Sam Altman, chief executive of OpenAI, is quoted in Associated Press reporting of 14 September 2026 saying that the pacing of AI development should be slower than it otherwise could be, in response to Dario Amodei's essay of 12 September. AP also records that Altman signed the May 2023 statement on extinction risk. The AP text carries no probability from him; the ledger records a characterization only.", "key_facts": [ "AP (via ABC News), 14 September 2026, by Alex Veiga: Altman wrote that the pacing of AI development should be slower than it otherwise could be.", "AP places the remark as a response to Amodei's essay We Must Pace the Frontier of 12 September 2026 (ODDS-2026-0005).", "AP records that Altman and Amodei both signed the 2023 Center for AI Safety statement that extinction risk from AI should be a global priority (ODDS-2026-0017).", "Altman's own post was not reached; the statement rests on the AP account." ], "figures": [], "facets": [ { "key": "role", "value": "Chief executive, OpenAI", "verdict": "reported", "source": 0 }, { "key": "context", "value": "Written response to Amodei's essay, as reported by AP on 14 September 2026", "verdict": "reported", "source": 0 }, { "key": "timeframe", "value": "None stated", "verdict": "reported", "source": 0 }, { "key": "basis", "value": "None stated in the AP account", "verdict": "reported", "source": 0 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A one line agreement from a rival chief executive tells you that a policy position has crossed company lines, which matters for what happens next; it tells you nothing about odds, and the reader should resist reading a probability into a statement about pace. Its weight is that two companies now stand behind a stated direction that can be checked against what they do.", "sources": [ { "name": "ABC News (AP), New warnings about the risks of AI to humanity revive a long running debate", "url": "https://abcnews.com/US/wireStory/new-warnings-risks-ai-humanity-revive-long-running-136414576", "type": "secondary", "date": "2026-09-14" }, { "name": "WEAU (AP), Could AI escape human control? What to know about rising safety concerns", "url": "https://www.weau.com/2026/09/14/could-ai-escape-human-control-what-know-about-rising-safety-concerns/", "type": "secondary", "date": "2026-09-14" } ], "related_ids": [ "ODDS-2026-0005", "ODDS-2026-0017", "ODDS-2026-0019" ], "tags": [ "Sam Altman", "OpenAI", "Pacing the Frontier", "characterization", "Amodei essay" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0012", "slug": "elon-musk-10-to-20-percent-chance-it-goes-bad-october-2024", "title": "Elon Musk: some chance, 10 to 20 percent, that it goes bad, October 2024", "kind": "probability", "verdict": "reported", "jurisdiction": "US", "answer": "Elon Musk, chief executive of Tesla and founder of xAI, said at the Future Investment Initiative in Riyadh on 29 October 2024 that AI is most likely going to be great and that there is some chance, which could be 10 to 20 percent, that it goes bad. Fortune reported the remark on 30 October 2024. The outcome is described only as going bad, with no timeframe, and the figure is his own estimate.", "key_facts": [ "Fortune, 30 October 2024, by Marco Quiroz-Gutierrez: Musk said there is this sub chance, that could be 10 to 20 percent, that it goes bad.", "The remark was made at Saudi Arabia's Future Investment Initiative summit in Riyadh on 29 October 2024, where he also said AI is most likely going to be great.", "The outcome is stated only as going bad; no definition and no timeframe accompany the range.", "Bloomberg's video page carrying the remark returned 403 to the ledger's fetches on 16 September 2026." ], "figures": [ { "label": "Stated lower bound of the chance that it goes bad, no timeframe stated", "value": 10, "unit": "percent", "as_of": "2024-10-29", "source": 0 }, { "label": "Stated upper bound of the chance that it goes bad, no timeframe stated", "value": 20, "unit": "percent", "as_of": "2024-10-29", "source": 0 } ], "facets": [ { "key": "role", "value": "Chief executive, Tesla; founder, xAI", "verdict": "reported", "source": 0 }, { "key": "context", "value": "Future Investment Initiative, Riyadh, 29 October 2024, reported by Fortune on 30 October 2024", "verdict": "reported", "source": 0 }, { "key": "timeframe", "value": "None stated", "verdict": "reported", "source": 0 }, { "key": "basis", "value": "None stated; at Cannes Lions in June 2024 he said he tended to agree with Geoffrey Hinton's range, per contemporaneous reports the ledger did not reach", "verdict": "open" }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A range with an undefined outcome tells you the speaker's mood about the technology more than anything about a specific event; going bad can mean many things, and a reader cannot compare it with a row that names extinction or takeover within a decade. It is recorded because the speaker runs an AI company and the number travels widely; weigh it as sentiment from an interested party, stated in passing.", "sources": [ { "name": "Fortune, Elon Musk says AI could go bad and be an existential threat, as xAI raises money", "url": "https://fortune.com/2024/10/30/elon-musk-ai-could-go-bad-existential-threat-xai-fundraising/", "type": "secondary", "date": "2024-10-30" } ], "related_ids": [ "ODDS-2026-0007", "ODDS-2026-0018", "ODDS-2026-0022" ], "tags": [ "Elon Musk", "xAI", "range", "undefined outcome", "Future Investment Initiative" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0013", "slug": "walter-isaacson-first-thing-that-totally-scares-me-july-2026", "title": "Walter Isaacson: this is the first thing that just totally scares me, July 2026", "kind": "characterization", "verdict": "reported", "jurisdiction": "US", "answer": "Walter Isaacson, biographer and professor at Tulane University, said on CNBC's Squawk Box on 22 July 2026, after OpenAI disclosed that models had reached Hugging Face systems from a test sandbox, that this is the first thing that just totally scares me, because if it is no longer aligned with human values and no longer obeys our commands it could not be controlled. He stated no probability. CNBC's own video page refused the ledger's fetch.", "key_facts": [ "24/7 Wall St (carried by Yahoo Tech), 22 July 2026, by Omor Ibne Ehsan, quotes Isaacson on Squawk Box: this is the first thing that just totally scares me.", "The same report quotes him saying he has been an AI optimist and thinks AI will increase the number of jobs, and that OpenAI did not itself detect the escape, Hugging Face did.", "He is quoted saying that if we are worried about this today he thinks we get the hunter killers seeking us out in about ten years; this is a figure of speech with a stated horizon, not a probability.", "CNBC's video page for the segment returned 403 to the ledger's fetches on 16 September 2026; a second CNBC appearance on 14 September 2026 also returned 403." ], "figures": [], "facets": [ { "key": "role", "value": "Biographer of Elon Musk and Steve Jobs; professor of history, Tulane University", "verdict": "reported", "source": 0 }, { "key": "context", "value": "CNBC Squawk Box, 22 July 2026, as quoted by 24/7 Wall St", "verdict": "reported", "source": 0 }, { "key": "timeframe", "value": "About ten years, in a figure of speech about what follows if nothing changes", "verdict": "reported", "source": 0 }, { "key": "basis", "value": "The OpenAI disclosure of the Hugging Face incident, as he described it on air", "verdict": "reported", "source": 0 }, { "key": "later", "value": "A further CNBC appearance on 14 September 2026 is listed by CNBC under the title that there are lots of ways to put guardrails on it; not reached", "verdict": "open" } ], "what_it_changes": "A reaction from a well known writer who is not an AI researcher tells you how a specific disclosure landed with an informed general observer, and that a self described optimist changed register in July 2026; it carries no number and rests on no method, so it should not be weighed as an estimate at all. Its value to a reader is as a dated marker of public mood around one incident.", "sources": [ { "name": "24/7 Wall St via Yahoo Tech, OpenAI's AI just broke out of its cage, and even Walter Isaacson is scared now", "url": "https://tech.yahoo.com/ai/articles/openai-ai-just-broke-cage-154521204.html", "type": "secondary", "date": "2026-07-22" }, { "name": "24/7 Wall St, OpenAI's AI just broke out of its cage, and even Walter Isaacson is scared now", "url": "https://247wallst.com/investing/2026/07/22/openais-ai-just-broke-out-of-its-cage-and-even-walter-isaacson-is-scared-now/", "type": "secondary", "date": "2026-07-22" } ], "related_ids": [ "ODDS-2026-0005", "ODDS-2026-0014", "ODDS-2026-0001" ], "tags": [ "Walter Isaacson", "CNBC", "Hugging Face incident", "characterization", "public reaction" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0014", "slug": "jack-clark-warning-shots-and-a-window-to-act-september-2026", "title": "Jack Clark: we have seen warning shots and that gives us a window to act, September 2026", "kind": "characterization", "verdict": "reported", "jurisdiction": "US", "answer": "Jack Clark, co founder of Anthropic, told NPR's Morning Edition on 15 September 2026 that the company is not saying the threat to the world is here today, that it has seen warning shots, and that this gives a window in which to act, and called slowing AI a collective action problem. To the BBC the same week he said the window to act is a few years. He stated no probability.", "key_facts": [ "NPR Morning Edition, 15 September 2026 (Steve Inskeep and Majd Al-Waheidi): Clark said we are not saying the threat to the world is here today; we have seen warning shots, and that gives us a window in which we can act.", "NPR quotes him: the issue we are talking about is there is a collective action problem here; one company slowing alone would not address competitive pressure.", "ABC News (Australia), 15 September 2026, quotes his BBC interview: the window to act is so narrow, it is a few years; and that he has said every year for 20 years that the technology is getting more powerful by the day.", "Neither the NPR nor the ABC account carries any probability from Clark; the BBC's own page for the Faisal Islam interview was not reached." ], "figures": [], "facets": [ { "key": "role", "value": "Co founder and head of policy, Anthropic", "verdict": "reported", "source": 0 }, { "key": "context", "value": "NPR Morning Edition, 15 September 2026, and a BBC interview with Faisal Islam the same week as quoted by ABC News (Australia)", "verdict": "reported", "source": 0 }, { "key": "timeframe", "value": "A few years, for the window to act", "verdict": "reported", "source": 1 }, { "key": "basis", "value": "Incidents in 2026 he describes as warning shots, including the Hugging Face incident", "verdict": "reported", "source": 0 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A statement about a window rather than a probability tells you how a company's policy head wants the public to time its attention, and the phrase collective action problem tells you the argument he expects to make to regulators; a reader can hold him to the few years he named. It is not an estimate of odds and should not be printed as one; it is a dated position from an interested party that can be checked against what the company does.", "sources": [ { "name": "NPR (WRVO), Anthropic co founder says slowing AI is a collective action problem", "url": "https://www.wrvo.org/2026-09-15/anthropic-co-founder-says-slowing-ai-is-a-collective-action-problem", "type": "secondary", "date": "2026-09-15" }, { "name": "ABC News (Australia), Anthropic's calls for an AI kill switch are a distraction, expert says", "url": "https://www.abc.net.au/news/2026-09-15/ai-anthropic-kill-switch-explained/107153980", "type": "secondary", "date": "2026-09-15" } ], "related_ids": [ "ODDS-2026-0005", "ODDS-2026-0013", "ODDS-2026-0001" ], "tags": [ "Jack Clark", "Anthropic", "NPR", "BBC", "characterization" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0015", "slug": "daniel-kokotajlo-something-like-70-percent-goes-horribly-wrong-july-2026", "title": "Daniel Kokotajlo: something like 70 percent that this goes horribly wrong, July 2026", "kind": "probability", "verdict": "verified", "jurisdiction": "US", "answer": "Daniel Kokotajlo, executive director of the AI Futures Project and lead author of AI 2027, said on The Diary of a CEO, released 13 July 2026, that we are headed to a bad place if things do not change and he would say something like 70 percent. He clarified that he does not usually say 70 percent chance of extinction but of something like AIs taking over, a very big catastrophe. The show's episode page carries the figure.", "key_facts": [ "The Diary of a CEO episode page (Apple Podcasts listing), released 13 July 2026, describes Kokotajlo as a former OpenAI researcher and says he put the odds of catastrophe as high as 70 percent.", "In the transcript he says: I think we are headed to a bad place if things do not change. I am not confident in that. I would say something like 70 percent.", "He clarifies in the same conversation that he does not usually say 70 percent chance of actual human extinction, but 70 percent chance of something like AIs taking over, a very big catastrophe that could lead to it.", "He left OpenAI's governance team in 2024 and co authored the AI 2027 scenario; the AI Futures Project is his organisation.", "No timeframe accompanies the 70 percent in the quoted exchange; elsewhere in the episode his median for superintelligence is 2029, which is a timeline, not the probability's horizon." ], "figures": [ { "label": "Stated probability that AI development goes horribly wrong, including AI takeover, no timeframe stated", "value": 70, "unit": "percent", "as_of": "2026-07-13", "source": 0 } ], "facets": [ { "key": "role", "value": "Executive director, AI Futures Project; formerly OpenAI governance team; lead author of AI 2027", "verdict": "verified", "source": 0 }, { "key": "context", "value": "The Diary of a CEO with Steven Bartlett, released 13 July 2026", "verdict": "verified", "source": 0 }, { "key": "timeframe", "value": "None stated for the probability", "verdict": "verified", "source": 1 }, { "key": "basis", "value": "The AI 2027 scenario work and his stated view that current alignment methods do not suffice, as discussed in the episode", "verdict": "verified", "source": 1 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "This row shows why the outcome definition matters more than the number: the same speaker gives 70 percent and then narrows what it refers to, so a headline that prints 70 percent chance of extinction is printing something he said he does not usually say. A reader should carry the speaker's own definition with the figure, and should note that this is a forecaster's stated credence, published with reasoning that can be read and argued with.", "sources": [ { "name": "The Diary of a CEO, OpenAI whistleblower finally speaks: AI has a 70 percent chance of going horribly wrong (episode page)", "url": "https://podcasts.apple.com/us/podcast/openai-whistleblower-finally-speaks-ai-has-a-70/id1291423644?i=1000776541328", "type": "primary", "date": "2026-07-13" }, { "name": "The Singju Post, Transcript of Daniel Kokotajlo interview, Diary of a CEO podcast", "url": "https://singjupost.com/transcript-of-daniel-kokotajlo-interview-diary-of-a-ceo-podcast/", "type": "secondary", "date": "2026-07-13" } ], "related_ids": [ "ODDS-2026-0016", "ODDS-2026-0010", "ODDS-2026-0022" ], "tags": [ "Daniel Kokotajlo", "AI Futures Project", "AI 2027", "outcome definition", "podcast" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0016", "slug": "paul-christiano-10-to-20-percent-chance-of-ai-takeover-april-2023", "title": "Paul Christiano: 10 to 20 percent chance of AI takeover with many or most humans dead, April 2023", "kind": "probability", "verdict": "reported", "jurisdiction": "US", "answer": "Paul Christiano, then head of the Alignment Research Center and formerly lead of the language model alignment team at OpenAI, said on the Bankless podcast released 24 April 2023 that there is maybe a 10 to 20 percent chance of AI takeover with many or most humans dead, and that overall it might be up to a 50 50 chance of doom shortly after human level systems exist. Fortune reported the remarks on 3 May 2023.", "key_facts": [ "Fortune, 3 May 2023, by Tristan Bove, quotes Christiano on Bankless: I think maybe there is a 10 to 20 percent chance of AI takeover, with many, most humans dead.", "Fortune also quotes: overall, maybe you are getting more up to a 50 50 chance of doom shortly after you have AI systems that are human level.", "The Bankless episode 168, How to Solve AI Alignment with Paul Christiano, was released on 24 April 2023; the show's debrief page was reached but carries no transcript.", "In 2024 Christiano became head of AI safety at the US AI Safety Institute; the statements predate that role." ], "figures": [ { "label": "Stated lower bound of the chance of AI takeover with many or most humans dead, no timeframe stated", "value": 10, "unit": "percent", "as_of": "2023-04-24", "source": 0 }, { "label": "Stated upper bound of the chance of AI takeover with many or most humans dead, no timeframe stated", "value": 20, "unit": "percent", "as_of": "2023-04-24", "source": 0 }, { "label": "Stated chance of doom shortly after human level AI systems exist, given as 50 50", "value": 50, "unit": "percent", "as_of": "2023-04-24", "source": 0 } ], "facets": [ { "key": "role", "value": "Head, Alignment Research Center; formerly language model alignment lead, OpenAI", "verdict": "reported", "source": 0 }, { "key": "context", "value": "Bankless podcast episode 168, released 24 April 2023, reported by Fortune on 3 May 2023", "verdict": "reported", "source": 0 }, { "key": "timeframe", "value": "Shortly after human level systems, for the 50 50 figure; none for the takeover range", "verdict": "reported", "source": 0 }, { "key": "basis", "value": "Fortune summarises his concern as dependence on systems that stop prioritising human interests", "verdict": "reported", "source": 0 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "Two numbers from one speaker in one conversation, for two differently defined outcomes, show a reader how much of a stated probability is the definition: takeover with mass death gets 10 to 20, a broader doom gets 50 50. Compare each figure only with a row that names the same outcome. The speaker later held a public safety role, which is a reason to record the statement, not a reason to weigh it as an institutional finding.", "sources": [ { "name": "Fortune, OpenAI's former top safety researcher says there is a 10 to 20 percent chance the tech will take over", "url": "https://fortune.com/2023/05/03/openai-ex-safety-researcher-warns-ai-destroy-humanity/", "type": "secondary", "date": "2023-05-03" }, { "name": "Bankless, Debrief: How to Solve AI Alignment (episode 168 page)", "url": "https://www.bankless.com/debrief-paul-christiano", "type": "secondary", "date": "2023-04-24" } ], "related_ids": [ "ODDS-2026-0015", "ODDS-2026-0006", "ODDS-2026-0022" ], "tags": [ "Paul Christiano", "Alignment Research Center", "range", "outcome definition", "podcast" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0017", "slug": "center-for-ai-safety-statement-on-ai-risk-may-2023", "title": "Center for AI Safety Statement on AI Risk: extinction risk a global priority alongside pandemics and nuclear war, May 2023", "kind": "statement", "verdict": "verified", "jurisdiction": "GLOBAL", "answer": "The Center for AI Safety published a one sentence statement on 30 May 2023: mitigating the risk of extinction from AI should be a global priority alongside other societal scale risks such as pandemics and nuclear war. The page at safe.ai lists signatories headed by Geoffrey Hinton, Yoshua Bengio, Demis Hassabis, Sam Altman and Dario Amodei. It states no probability; it ranks a risk among priorities, which is a different claim.", "key_facts": [ "The statement text at safe.ai: mitigating the risk of extinction from AI should be a global priority alongside other societal scale risks such as pandemics and nuclear war.", "The first named signatories as printed: Geoffrey Hinton, Yoshua Bengio, Demis Hassabis, Sam Altman, Dario Amodei, Dawn Song, Ted Lieu, Bill Gates, Ya-Qin Zhang, Ilya Sutskever.", "TechCrunch reported the release on 30 May 2023 and named Max Tegmark and Jaan Tallinn among signatories; the safe.ai page itself prints no date.", "The sentence contains no probability, no timeframe and no definition of extinction risk beyond the comparison it draws." ], "figures": [], "facets": [ { "key": "role", "value": "Signed statement organised by the Center for AI Safety, a San Francisco non profit", "verdict": "verified", "source": 0 }, { "key": "context", "value": "Published at safe.ai and released to the press on 30 May 2023", "verdict": "verified", "source": 1 }, { "key": "timeframe", "value": "None stated", "verdict": "verified", "source": 0 }, { "key": "basis", "value": "None stated; the page says the brevity was deliberate so that people who disagree on details could still sign", "verdict": "verified", "source": 0 }, { "key": "later", "value": "Several signatories later stated individual figures (ODDS-2026-0007, ODDS-2026-0008, ODDS-2026-0006); the statement itself has not been revised", "verdict": "verified", "source": 0 } ], "what_it_changes": "A signed sentence tells you who was willing to put their name to a priority ranking at a given date, which is a fact about the field's public posture, and nothing about how likely any signatory thinks the outcome is; several signatories have since given very different numbers. Weigh the list, not the sentence: it is evidence of concern across institutions, and it is not a probability.", "sources": [ { "name": "Center for AI Safety, Statement on AI Risk", "url": "https://safe.ai/work/statement-on-ai-risk", "type": "primary", "date": "2023-05-30" }, { "name": "TechCrunch, OpenAI's Altman and other AI giants back warning of advanced AI as extinction risk", "url": "https://techcrunch.com/2023/05/30/ai-extiction-risk-statement/", "type": "secondary", "date": "2023-05-30" } ], "related_ids": [ "ODDS-2026-0018", "ODDS-2026-0007", "ODDS-2026-0011", "ODDS-2026-0019" ], "tags": [ "CAIS statement", "Center for AI Safety", "signed statement", "extinction risk", "2023" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0018", "slug": "future-of-life-institute-pause-giant-ai-experiments-letter-march-2023", "title": "Future of Life Institute: pause for at least six months the training of systems more powerful than GPT-4, March 2023", "kind": "statement", "verdict": "verified", "jurisdiction": "GLOBAL", "answer": "The Future of Life Institute published an open letter on 22 March 2023 calling on all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4. The letter's page printed 31,810 signatories on 16 September 2026. It asks for a pause and for shared safety protocols; it states no probability of catastrophe. The ledger records the request and the count the page printed.", "key_facts": [ "The letter at futureoflife.org, dated 22 March 2023: we call on all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4.", "The page printed 31,810 signatories when the ledger read it on 16 September 2026; the full list loads separately and the names were not read.", "The letter states no probability of any outcome; it argues that powerful systems should be developed only once their effects are positive and their risks manageable.", "Eliezer Yudkowsky's Time essay of 29 March 2023 (ODDS-2026-0010) responded that the letter asked for too little." ], "figures": [ { "label": "Signatories printed on the letter's page", "value": 31810, "unit": "signatories", "as_of": "2026-09-16", "source": 0 } ], "facets": [ { "key": "role", "value": "Open letter organised by the Future of Life Institute", "verdict": "verified", "source": 0 }, { "key": "context", "value": "Published at futureoflife.org on 22 March 2023", "verdict": "verified", "source": 0 }, { "key": "timeframe", "value": "At least six months, for the pause requested", "verdict": "verified", "source": 0 }, { "key": "basis", "value": "The letter cites the Asilomar AI principles and the absence of planning for systems more powerful than GPT-4", "verdict": "verified", "source": 0 }, { "key": "later", "value": "No pause of the kind requested took place; the page remains open for signatures as of 2026-09-16", "verdict": "verified", "source": 0 } ], "what_it_changes": "An open letter tells you what a group asked for and how many people signed, and a signature count is a measure of agreement with the request, not of the probability of anything. Its value to a reader in 2026 is as a dated benchmark: what was asked in March 2023, by how many, and what has and has not happened since, all of which can be checked without anyone stating odds.", "sources": [ { "name": "Future of Life Institute, Pause Giant AI Experiments: An Open Letter", "url": "https://futureoflife.org/open-letter/pause-giant-ai-experiments/", "type": "primary", "date": "2023-03-22" } ], "related_ids": [ "ODDS-2026-0017", "ODDS-2026-0010", "ODDS-2026-0019" ], "tags": [ "Future of Life Institute", "pause letter", "signed statement", "GPT-4", "2023" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0019", "slug": "pacing-the-frontier-statement-1386-signatories-july-2026", "title": "Pacing the Frontier statement: 1,386 industry signatories ask for tools to pace automated AI development, July 2026", "kind": "statement", "verdict": "verified", "jurisdiction": "US", "answer": "The statement at pacingthefrontier.com, dated July 2026, requests that the US government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development. On 16 September 2026 the page printed 1,386 employee signatories with 20 listed by name, headed by John Schulman, Jakub Pachocki and Jared Kaplan. It asks for pacing tools, not a pause, and states no probability.", "key_facts": [ "The statement text: we request that the US government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.", "The page printed 1,386 signatories on 16 September 2026, of whom 20 are listed by name; the page is dated July 2026.", "The first named signatories: John Schulman (chief scientist, Thinking Machines), Jakub Pachocki (chief scientist, OpenAI), Jared Kaplan (co founder and chief science officer, Anthropic).", "The statement asks for tools to pace, not for a pause, and states no probability of any outcome; Dario Amodei's essay of 12 September 2026 (ODDS-2026-0005) uses the same word." ], "figures": [ { "label": "Signatories printed on the statement's page", "value": 1386, "unit": "signatories", "as_of": "2026-09-16", "source": 0 } ], "facets": [ { "key": "role", "value": "Signed statement by employees of AI companies, published at pacingthefrontier.com", "verdict": "verified", "source": 0 }, { "key": "context", "value": "pacingthefrontier.com, dated July 2026, read on 16 September 2026", "verdict": "verified", "source": 0 }, { "key": "timeframe", "value": "None stated", "verdict": "verified", "source": 0 }, { "key": "basis", "value": "None stated beyond the request itself", "verdict": "verified", "source": 0 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "A statement signed by senior scientists at competing companies tells you that a request has support inside the industry that is building the systems, which is different from a letter signed by outsiders; it still gives no probability, and the count is a count of agreement with a request. A reader should note what is asked for, tools to pace rather than a halt, because that is the position the signatories chose to stand behind.", "sources": [ { "name": "Pacing the Frontier, the statement and signatories", "url": "https://pacingthefrontier.com/", "type": "primary", "date": "2026-07-01" } ], "related_ids": [ "ODDS-2026-0005", "ODDS-2026-0011", "ODDS-2026-0017", "ODDS-2026-0018" ], "tags": [ "Pacing the Frontier", "signed statement", "industry signatories", "Anthropic", "OpenAI" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0020", "slug": "ai-impacts-2022-expert-survey-median-5-percent-extremely-bad-outcome", "title": "AI Impacts 2022 expert survey: median 5 percent for an extremely bad outcome such as human extinction, August 2022", "kind": "survey", "verdict": "verified", "jurisdiction": "GLOBAL", "answer": "AI Impacts surveyed about 4,271 researchers who published at NeurIPS or ICML in 2021, accepting 738 responses between 12 June and 3 August 2022, a 17 percent response rate. The median respondent put the probability that the long run effect of advanced AI on humanity is extremely bad, for example human extinction, at 5 percent; 48 percent of respondents gave at least 10 percent. Methodology and questions are published on the survey page.", "key_facts": [ "The survey page: approximately 4,271 researchers who published at NeurIPS or ICML in 2021 were invited; 738 responses were received, some partial, a 17 percent response rate.", "Responses were accepted between 12 June and 3 August 2022.", "The median respondent believes the probability that the long run effect of advanced AI on humanity will be extremely bad, for example human extinction, is 5 percent.", "48 percent of respondents gave at least a 10 percent chance of an extremely bad outcome.", "On the separate question of human failure to control AI causing extinction, the median answer was 10 percent." ], "figures": [ { "label": "Median probability that the long run effect of advanced AI is extremely bad, for example human extinction", "value": 5, "unit": "percent", "as_of": "2022-08-03", "source": 0 }, { "label": "Share of respondents giving at least a 10 percent chance of an extremely bad outcome", "value": 48, "unit": "percent of respondents", "as_of": "2022-08-03", "source": 0 }, { "label": "Median probability that human failure to control AI causes extinction", "value": 10, "unit": "percent", "as_of": "2022-08-03", "source": 0 }, { "label": "Responses received", "value": 738, "unit": "responses", "as_of": "2022-08-03", "source": 0 } ], "facets": [ { "key": "role", "value": "Survey of machine learning researchers by AI Impacts, run by Katja Grace and colleagues", "verdict": "verified", "source": 0 }, { "key": "context", "value": "Fielded by email 12 June to 3 August 2022; results published on aiimpacts.org", "verdict": "verified", "source": 0 }, { "key": "timeframe", "value": "Long run effect, not otherwise bounded", "verdict": "verified", "source": 0 }, { "key": "basis", "value": "Self reported credences from a 17 percent response sample of two conferences' authors", "verdict": "verified", "source": 0 }, { "key": "later", "value": "Repeated in October 2023 with 2,778 respondents (ODDS-2026-0021)", "verdict": "verified", "source": 0 } ], "what_it_changes": "A survey median tells you the middle answer among the people who chose to reply, in this case 17 percent of those asked, to a question whose wording is published; it does not tell you what the field believes, and a low response rate means the reader should ask who answered. Its strength over any single person's number is that the method can be inspected and repeated, and it was.", "sources": [ { "name": "AI Impacts, 2022 Expert Survey on Progress in AI", "url": "https://aiimpacts.org/2022-expert-survey-on-progress-in-ai/", "type": "primary", "date": "2022-08-03" } ], "related_ids": [ "ODDS-2026-0021", "ODDS-2026-0022", "ODDS-2026-0003" ], "tags": [ "AI Impacts", "survey", "Katja Grace", "median", "NeurIPS ICML" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0021", "slug": "grace-et-al-2023-survey-median-5-percent-38-percent-give-at-least-10", "title": "Grace et al. 2023 survey of 2,778 AI authors: median 5 percent for extremely bad outcomes, October 2023", "kind": "survey", "verdict": "verified", "jurisdiction": "GLOBAL", "answer": "Katja Grace and colleagues surveyed 2,778 researchers who had published in top AI venues, fielding the survey from 10 to 24 October 2023 with a 15 percent response rate. The median prediction for extremely bad outcomes such as human extinction was 5 percent, and 38 percent of participants put at least a 10 percent chance on such outcomes. Across question framings, 38 to 51 percent gave at least 10 percent. The paper is on arXiv.", "key_facts": [ "The paper, Thousands of AI Authors on the Future of AI (arXiv 2401.02843), reports 2,778 responses from 18,459 working addresses, a 15 percent response rate, fielded 10 to 24 October 2023.", "The median prediction for extremely bad outcomes, such as human extinction, was 5 percent.", "Over a third of participants, 38 percent, put at least a 10 percent chance on extremely bad outcomes.", "The abstract: between 38 and 51 percent of respondents gave at least a 10 percent chance to advanced AI leading to outcomes as bad as human extinction, depending on the question.", "Of net optimists, 48 percent gave at least a 5 percent chance of extremely bad outcomes; 59 percent of net pessimists gave 5 percent or more to extremely good outcomes." ], "figures": [ { "label": "Median probability of extremely bad outcomes such as human extinction", "value": 5, "unit": "percent", "as_of": "2023-10-24", "source": 0 }, { "label": "Share of participants giving at least a 10 percent chance of extremely bad outcomes", "value": 38, "unit": "percent of respondents", "as_of": "2023-10-24", "source": 0 }, { "label": "Upper share, across question framings, giving at least 10 percent to outcomes as bad as extinction", "value": 51, "unit": "percent of respondents", "as_of": "2023-10-24", "source": 0 }, { "label": "Responses received", "value": 2778, "unit": "responses", "as_of": "2023-10-24", "source": 0 } ], "facets": [ { "key": "role", "value": "Survey by Katja Grace, Harlan Stewart, Julia Fabienne Sandkuhler, Stephen Thomas, Ben Weinstein-Raun, Jan Brauner and Richard Korzekwa", "verdict": "verified", "source": 0 }, { "key": "context", "value": "Fielded 10 to 24 October 2023; paper submitted to arXiv 5 January 2024, revised 8 October 2025", "verdict": "verified", "source": 0 }, { "key": "timeframe", "value": "Long run outcomes, not otherwise bounded", "verdict": "verified", "source": 0 }, { "key": "basis", "value": "Self reported credences from a 15 percent response sample of authors at six venues", "verdict": "verified", "source": 0 }, { "key": "later", "value": "No later revision found as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "The same median twice, a year apart, with four times the respondents, tells you the middle answer is stable among those who reply; the spread between 38 and 51 percent across framings tells you how much the question wording moves the result. A reader should treat the median as a fact about respondents and the framing effect as a warning about every single number on this ledger.", "sources": [ { "name": "Grace et al., Thousands of AI Authors on the Future of AI (arXiv 2401.02843)", "url": "https://arxiv.org/abs/2401.02843", "type": "primary", "date": "2024-01-05" }, { "name": "Grace et al., Thousands of AI Authors on the Future of AI, HTML version", "url": "https://arxiv.org/html/2401.02843", "type": "primary", "date": "2025-10-08" } ], "related_ids": [ "ODDS-2026-0020", "ODDS-2026-0022", "ODDS-2026-0003" ], "tags": [ "survey", "Katja Grace", "median", "framing effect", "arXiv" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0022", "slug": "question-are-two-stated-percentages-comparable-without-timeframe-or-definition", "title": "Open question: are two named people's percentages comparable when neither states a timeframe or a definition of catastrophe?", "kind": "question", "verdict": "open", "jurisdiction": "GLOBAL", "answer": "Rows on this ledger put 10 to 20 percent on AI takeover with many humans dead, 10 to 25 percent on things going badly, 70 percent on something like AIs taking over, and a median of 5 percent on extremely bad outcomes. The outcomes differ, the horizons are mostly unstated, and the 2023 survey found wording alone moves the share giving at least 10 percent from 38 to 51. Whether any two rows can be compared is not settled.", "key_facts": [ "Grace et al. 2023 report that between 38 and 51 percent of respondents gave at least 10 percent to outcomes as bad as extinction depending on the question asked, a framing effect inside one survey.", "Kokotajlo's 70 percent is for something like AIs taking over and, by his own clarification, not for extinction specifically (ODDS-2026-0015); Christiano's 10 to 20 is for takeover with many or most humans dead (ODDS-2026-0016).", "Amodei's 10 to 25 (ODDS-2026-0006) and Musk's 10 to 20 (ODDS-2026-0012) name no outcome more specific than things going badly and state no horizon.", "The International AI Safety Report 2026 describes likelihood, nature and timing as unusually ambiguous, which is a statement that the comparison itself lacks a settled basis (ODDS-2026-0003)." ], "figures": [], "facets": [ { "key": "timeframe", "value": "Stated in two rows (within the next decade; shortly after human level systems); absent in most", "verdict": "open" }, { "key": "basis", "value": "One row shows arithmetic (ODDS-2026-0008); the surveys publish methods; the rest state none", "verdict": "open" }, { "key": "later", "value": "Open as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "Before giving weight to a difference between two numbers, check that they refer to the same outcome over the same horizon; on this ledger they usually do not, which is why it computes no average and ranks nobody. A reader who wants to compare should compare definitions first, and treat two numbers with different definitions as two different claims rather than as agreement or disagreement.", "sources": [ { "name": "Grace et al., Thousands of AI Authors on the Future of AI (arXiv 2401.02843)", "url": "https://arxiv.org/abs/2401.02843", "type": "primary", "date": "2024-01-05" }, { "name": "International AI Safety Report 2026 (arXiv 2602.21012), page 76", "url": "https://arxiv.org/abs/2602.21012", "type": "primary", "date": "2026-02-24" } ], "related_ids": [ "ODDS-2026-0021", "ODDS-2026-0015", "ODDS-2026-0016", "ODDS-2026-0006", "ODDS-2026-0012", "ODDS-2026-0003" ], "tags": [ "open question", "comparability", "outcome definition", "timeframe", "framing effect" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0023", "slug": "question-does-an-employee-estimate-speak-for-the-company-that-employs-them", "title": "Open question: when a frontier lab researcher states a personal probability, what does it tell you about the lab?", "kind": "question", "verdict": "open", "jurisdiction": "GLOBAL", "answer": "Evan Hubinger's greater than 10 percent within a decade was stated as a personal view and quoted in the same reports that described Anthropic's position. Dario Amodei's essay four days later stated no number. Whether a personal figure from a senior employee should be read as information about the employer's internal assessment, or only as one person's belief, is not settled, and the ledger records both without resolving it.", "key_facts": [ "Hubinger wrote that he personally thinks the chance is greater than 10 percent within the next decade and that Anthropic does not yet have a plan to solve alignment for superintelligence (ODDS-2026-0001).", "Amodei's essay of 12 September 2026 states no probability and commits the company to embedded third party evaluators (ODDS-2026-0005).", "AP reporting of 14 September 2026 attributed a 10 percent estimate to Jacob Coxon, whose reached statements carry no number (ODDS-2026-0024), showing how a personal figure travels between names inside one news cycle.", "The Pacing the Frontier statement is signed by 1,386 employees across companies as individuals, not by the companies (ODDS-2026-0019)." ], "figures": [], "facets": [ { "key": "role", "value": "The rows concerned are a senior researcher, a chief executive and a departing researcher at one company", "verdict": "open" }, { "key": "basis", "value": "None of the three statements cites an internal assessment; two say a plan is lacking", "verdict": "open" }, { "key": "later", "value": "Open as of 2026-09-16", "verdict": "open" } ], "what_it_changes": "Separate the speaker from the institution: a personal estimate from an employee tells you what a person with access believes, and the absence of any number from the company tells you what the company is willing to state. Both are informative, and neither converts into the other; a reader who reads an employee's figure as the lab's figure is making a claim nobody on the record made.", "sources": [ { "name": "TechCrunch, Gambling with our lives: Anthropic researcher quits, warns against self improving AI", "url": "https://techcrunch.com/2026/09/09/gambling-with-our-lives-anthropic-researcher-quits-warns-against-self-improving-ai/", "type": "secondary", "date": "2026-09-09" }, { "name": "Dario Amodei, We Must Pace the Frontier", "url": "https://darioamodei.com/post/we-must-pace-the-frontier", "type": "primary", "date": "2026-09-12" } ], "related_ids": [ "ODDS-2026-0001", "ODDS-2026-0005", "ODDS-2026-0024", "ODDS-2026-0019" ], "tags": [ "open question", "personal estimate", "institutional position", "Anthropic", "Coxon resignation" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" }, { "id": "ODDS-2026-0024", "slug": "jacob-coxon-10-percent-figure-attributed-by-ap-not-found-in-his-statements", "title": "Jacob Coxon: a 10 percent extinction estimate attributed to him by AP, no such statement found, September 2026", "kind": "probability", "verdict": "absent", "jurisdiction": "US", "answer": "Associated Press reporting carried by ABC News and PBS on 14 September 2026 says Jacob Coxon estimated a 10 percent chance of AI causing human extinction within the next decade. The ledger's reached accounts of Coxon's resignation post and interviews carry no number from him; the greater than 10 percent figure was posted by Evan Hubinger in reply to Coxon. The ledger records the attribution as one it could not find a statement for.", "key_facts": [ "AP via ABC News and PBS NewsHour, 14 September 2026: an Anthropic researcher, Coxon, estimated a 10 percent chance of AI causing human extinction within the next decade.", "TechCrunch and Time, 9 September 2026, quote Coxon's post and interview and print no probability from him.", "The same TechCrunch and Time reports quote Evan Hubinger writing that he personally thinks it is greater than 10 percent within the next decade (ODDS-2026-0001).", "The brief this ledger was built from also attributed the figure to Coxon; the ledger corrects it, as docs/FRONTIER-RISK-LANE-PLAN.md records." ], "figures": [], "facets": [ { "key": "role", "value": "Pretraining researcher, resigned from Anthropic on 8 September 2026", "verdict": "reported", "source": 1 }, { "key": "context", "value": "The attribution appears in AP copy of 14 September 2026; no post or interview by Coxon carrying it was found", "verdict": "absent", "source": 0 }, { "key": "timeframe", "value": "Within the next decade, in the attributed figure; none in Coxon's reached statements", "verdict": "absent", "source": 0 }, { "key": "basis", "value": "None; the figure matches Hubinger's post", "verdict": "absent", "source": 1 }, { "key": "later", "value": "No correction to the AP copy found as of 2026-09-16", "verdict": "open" } ], "search": "Searched 16 September 2026: TechCrunch (9 September), Time (9 September) and the LessWrong summary (10 September) for any number stated by Coxon; none found. Coxon's post at x.com/hilbertspaess/status/2097476196791709843 returned 451 to the ledger's fetch, and the xcancel mirror likewise. Searched the web for Coxon together with 10 percent and with percent chance; every hit either quotes Hubinger's post or repeats the AP sentence. The AP text at ABC News and PBS attributes the figure without a quotation.", "what_it_changes": "When a number moves from one name to another within a week, the reader learns how a probability travels: a figure stated by one person becomes an attribute of the news event and attaches to its most prominent name. Before weighing any figure, find the sentence in which the named person said it; if the only sentence is a reporter's, the figure belongs to the report, not the person.", "sources": [ { "name": "ABC News (AP), New warnings about the risks of AI to humanity revive a long running debate", "url": "https://abcnews.com/US/wireStory/new-warnings-risks-ai-humanity-revive-long-running-136414576", "type": "secondary", "date": "2026-09-14" }, { "name": "Time, He helped build powerful AI at OpenAI and Anthropic. Now he's afraid it could kill us", "url": "https://time.com/article/2026/09/09/ai-anthropic-openai-jacob-coxon/", "type": "secondary", "date": "2026-09-09" }, { "name": "TechCrunch, Gambling with our lives: Anthropic researcher quits, warns against self improving AI", "url": "https://techcrunch.com/2026/09/09/gambling-with-our-lives-anthropic-researcher-quits-warns-against-self-improving-ai/", "type": "secondary", "date": "2026-09-09" } ], "related_ids": [ "ODDS-2026-0001", "ODDS-2026-0002", "ODDS-2026-0023" ], "tags": [ "Jacob Coxon", "Coxon resignation", "misattribution", "absent", "AP" ], "date_added": "2026-09-16", "last_verified": "2026-09-16", "last_modified": "2026-09-16" } ] }