{ "generated": "2026-11-07T15:39:52.499230+00:00", "date": "2026-11-07", "formula_version": "1.0", "weights": { "agi": 0.3, "pdoom": 0.25, "crowd": 0.15, "incidents": 0.15, "safety": 0.15 }, "score": 61.6, "minutes_to_midnight": 23.0, "change_1d": -0.1, "change_7d": 1.2, "components": { "agi": { "score": 46.9, "weight": 0.3, "contribution": 14.06 }, "crowd": { "score": 34.4, "weight": 0.15, "contribution": 5.16 }, "pdoom": { "score": 75.0, "weight": 0.25, "contribution": 18.75 }, "incidents": { "score": 89.5, "weight": 0.15, "contribution": 13.42 }, "safety": { "score": 68.1, "weight": 0.15, "contribution": 10.22 } }, "movers": [ { "component": "agi", "delta_points": -0.31, "from": 47.9, "to": 46.9 }, { "component": "incidents", "delta_points": 0.19, "from": 88.2, "to": 89.5 }, { "component": "safety", "delta_points": 0.01, "from": 68.1, "to": 68.1 }, { "component": "crowd", "delta_points": 0.0, "from": 34.4, "to": 34.4 }, { "component": "pdoom", "delta_points": 0.0, "from": 75.0, "to": 75.0 } ], "input_moves": [ { "label": "Will we get AGI before 2028?", "venue": "Manifold", "now": 0.29000000000000015, "d1": null, "d7": null }, { "label": "Any company announces AGI before Jan 1, 2028", "venue": "Kalshi", "now": 0.485, "d1": null, "d7": null }, { "label": "OpenAI announces it has achieved AGI before 2027", "venue": "Polymarket", "now": 0.07, "d1": null, "d7": null }, { "label": "Will we get AGI before 2030?", "venue": "Manifold", "now": 0.5166313020593701, "d1": null, "d7": null }, { "label": "Any company announces AGI before Jan 1, 2030", "venue": "Kalshi", "now": 0.795, "d1": null, "d7": null }, { "label": "Massive catastrophe caused by AI before 2030?", "venue": "Manifold", "now": 0.3911000869467075, "d1": null, "d7": null }, { "label": "AI kills at least 10% of humans before 2100?", "venue": "Manifold", "now": 0.2970283617502112, "d1": null, "d7": null } ], "markets": { "agi": [ { "id": "manifold-agi-2028", "venue": "Manifold", "kind": "manifold", "slug": "will-we-get-agi-before-2028-ff560f9e9346", "bucket": "near", "label": "Will we get AGI before 2028?", "url": "https://manifold.markets/RemNi/will-we-get-agi-before-2028-ff560f9e9346", "traders": 347, "prob": 0.29000000000000015, "ok": true }, { "id": "kalshi-agi-2028", "venue": "Kalshi", "kind": "kalshi", "series": "KXAGICO", "ticker": "KXAGICO-COMP-27Q4", "bucket": "near", "label": "Any company announces AGI before Jan 1, 2028", "url": "https://kalshi.com/markets/kxagico", "open_interest": "10263.19", "prob": 0.485, "ok": true }, { "id": "polymarket-openai-agi-2027", "venue": "Polymarket", "kind": "polymarket", "slug": "openai-announces-it-has-achieved-agi-before-2027", "bucket": "near", "label": "OpenAI announces it has achieved AGI before 2027", "url": "https://polymarket.com/event/openai-announces-it-has-achieved-agi-before-2027", "volume": "365837.888908", "prob": 0.07, "ok": true }, { "id": "manifold-agi-2030", "venue": "Manifold", "kind": "manifold", "slug": "will-we-get-agi-before-2030", "bucket": "far", "label": "Will we get AGI before 2030?", "url": "https://manifold.markets/RemNi/will-we-get-agi-before-2030", "traders": 351, "prob": 0.5166313020593701, "ok": true }, { "id": "kalshi-agi-2030", "venue": "Kalshi", "kind": "kalshi", "series": "KXAGICO", "ticker": "KXAGICO-COMP-29", "bucket": "far", "label": "Any company announces AGI before Jan 1, 2030", "url": "https://kalshi.com/markets/kxagico", "open_interest": "3025.99", "prob": 0.795, "ok": true } ], "crowd": [ { "id": "manifold-catastrophe-2030", "venue": "Manifold", "kind": "manifold", "slug": "will-there-be-a-massive-catastrophe", "label": "Massive catastrophe caused by AI before 2030?", "url": "https://manifold.markets/ersatz/will-there-be-a-massive-catastrophe", "traders": 48, "prob": 0.3911000869467075, "ok": true }, { "id": "manifold-10pct-2100", "venue": "Manifold", "kind": "manifold", "slug": "global-catastrophe-series-by-2100-w-73fc5b6f55b0", "label": "AI kills at least 10% of humans before 2100?", "url": "https://manifold.markets/jack/global-catastrophe-series-by-2100-w-73fc5b6f55b0", "traders": 19, "prob": 0.2970283617502112, "ok": true } ] }, "pdoom": { "median": 37.5, "n_median": 21, "n_verified": 22, "rows": [ { "name": "Roman Yampolskiy", "slug": "roman-yampolskiy", "role": "AI safety researcher, University of Louisville", "camp": "researcher", "latest": { "date": "2024-03-12", "low": 99.999999, "high": 99.999999, "display": "99.999999%", "source_name": "X (@romanyam)", "source_url": "https://x.com/romanyam/status/1767575356155027503", "quote": "I am at 99.999999%", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" }, "mid": 99.999999, "statements": [ { "date": "2024-03-12", "low": 99.999999, "high": 99.999999, "display": "99.999999%", "source_name": "X (@romanyam)", "source_url": "https://x.com/romanyam/status/1767575356155027503", "quote": "I am at 99.999999%", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" } ], "excluded": false }, { "name": "Eliezer Yudkowsky", "slug": "eliezer-yudkowsky", "role": "Co-founder, Machine Intelligence Research Institute", "camp": "researcher", "latest": { "date": "2025-09-12", "low": 99.5, "high": 99.5, "display": "99.5%", "source_name": "The New York Times (Kevin Roose)", "source_url": "https://www.nytimes.com/2025/09/12/technology/ai-eliezer-yudkowsky-book.html", "quote": "The first time I met Eliezer Yudkowsky, he said there was a 99.5% chance that A.I. was going to kill me.", "quote_kind": "reported", "verified": true, "verified_on": "2026-10-21", "note": "Reported speech: the figure is relayed by the reporter, not quoted directly." }, "mid": 99.5, "statements": [ { "date": "2025-09-12", "low": 99.5, "high": 99.5, "display": "99.5%", "source_name": "The New York Times (Kevin Roose)", "source_url": "https://www.nytimes.com/2025/09/12/technology/ai-eliezer-yudkowsky-book.html", "quote": "The first time I met Eliezer Yudkowsky, he said there was a 99.5% chance that A.I. was going to kill me.", "quote_kind": "reported", "verified": true, "verified_on": "2026-10-21", "note": "Reported speech: the figure is relayed by the reporter, not quoted directly." } ], "excluded": false }, { "name": "Dan Hendrycks", "slug": "dan-hendrycks", "role": "Director, Center for AI Safety", "camp": "researcher", "latest": { "date": "2023-04-02", "low": 80, "high": 80, "display": ">80%", "source_name": "X (@DanHendrycks)", "source_url": "https://x.com/DanHendrycks/status/1642394635657162753", "quote": "As it happens, my p(doom) > 80%, but it has been lower in the past. Two years ago it was ~20%.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Stated as a lower bound; counted as 80%." }, "mid": 80.0, "statements": [ { "date": "2023-04-02", "low": 80, "high": 80, "display": ">80%", "source_name": "X (@DanHendrycks)", "source_url": "https://x.com/DanHendrycks/status/1642394635657162753", "quote": "As it happens, my p(doom) > 80%, but it has been lower in the past. Two years ago it was ~20%.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Stated as a lower bound; counted as 80%." } ], "excluded": false }, { "name": "Daniel Kokotajlo", "slug": "daniel-kokotajlo", "role": "Executive Director, AI Futures Project; former OpenAI researcher", "camp": "researcher", "latest": { "date": "2026-07-13", "low": 70, "high": 70, "display": "70%", "source_name": "The Diary Of A CEO (Steven Bartlett)", "source_url": "https://podcastrex.com/shows/the-diary-of-a-ceo-with-steven-bartlett/openai-whistleblower-finally-speaks-ai-has-a-70-chance-of-going-horribly-wrong/transcript", "quote": "\u2026it's possible that we'll end up essentially creating a new species that ends up ruling the world with a 70% chance that this goes horribly wrong", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "From an automated episode transcript. He frames 70% as things 'going horribly wrong' (AI takeover or comparable catastrophe), not strictly extinction." }, "mid": 70.0, "statements": [ { "date": "2026-07-13", "low": 70, "high": 70, "display": "70%", "source_name": "The Diary Of A CEO (Steven Bartlett)", "source_url": "https://podcastrex.com/shows/the-diary-of-a-ceo-with-steven-bartlett/openai-whistleblower-finally-speaks-ai-has-a-70-chance-of-going-horribly-wrong/transcript", "quote": "\u2026it's possible that we'll end up essentially creating a new species that ends up ruling the world with a 70% chance that this goes horribly wrong", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "From an automated episode transcript. He frames 70% as things 'going horribly wrong' (AI takeover or comparable catastrophe), not strictly extinction." } ], "excluded": false }, { "name": "Zvi Mowshowitz", "slug": "zvi-mowshowitz", "role": "Writer, Don't Worry About the Vase", "camp": "commentator", "latest": { "date": "2025-09-06", "low": 70, "high": 70, "display": "70%", "source_name": "The Cognitive Revolution podcast", "source_url": "https://www.cognitiverevolution.ai/zvi-mowshowitz-on-longer-timelines-rl-induced-doom-and-why-china-is-refusing-h20s/", "quote": "[Host: \u201cSo last time we spoke, I believe your p(doom) was 70%. Has it ticked down at all?\u201d] \u2026 I don't think it changed substantially, though.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "He confirms the host's 70% rather than restating the number." }, "mid": 70.0, "statements": [ { "date": "2025-09-06", "low": 70, "high": 70, "display": "70%", "source_name": "The Cognitive Revolution podcast", "source_url": "https://www.cognitiverevolution.ai/zvi-mowshowitz-on-longer-timelines-rl-induced-doom-and-why-china-is-refusing-h20s/", "quote": "[Host: \u201cSo last time we spoke, I believe your p(doom) was 70%. Has it ticked down at all?\u201d] \u2026 I don't think it changed substantially, though.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "He confirms the host's 70% rather than restating the number." } ], "excluded": false }, { "name": "Emad Mostaque", "slug": "emad-mostaque", "role": "Founder, Stability AI", "camp": "industry", "latest": { "date": "2024-12-04", "low": 50, "high": 50, "display": "50%", "source_name": "X (@EMostaque)", "source_url": "https://x.com/EMostaque/status/1864266899170767105", "quote": "My P(doom) is 50%. Given an undefined time period the probability of systems that are more capable than humans \u2026 wiping us all out is a coin toss", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" }, "mid": 50.0, "statements": [ { "date": "2024-12-04", "low": 50, "high": 50, "display": "50%", "source_name": "X (@EMostaque)", "source_url": "https://x.com/EMostaque/status/1864266899170767105", "quote": "My P(doom) is 50%. Given an undefined time period the probability of systems that are more capable than humans \u2026 wiping us all out is a coin toss", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" } ], "excluded": false }, { "name": "Holden Karnofsky", "slug": "holden-karnofsky", "role": "Co-founder, Open Philanthropy", "camp": "researcher", "latest": { "date": "2022-08-29", "low": 10, "high": 90, "display": "10\u201390%", "source_name": "AI Alignment Forum", "source_url": "https://www.alignmentforum.org/posts/rCJQAkPTEypGjSJ8X/how-might-we-align-transformative-ai-if-it-s-developed-very", "quote": "I don't currently have much sympathy for someone who's highly confident that AI takeover would or would not happen (that is, for anyone who thinks the odds of AI takeover in this nearcast are under 10% or over 90%).", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Caveat: this is the range of views he considers reasonable, not a personal point estimate. Counted at the 50% midpoint." }, "mid": 50.0, "statements": [ { "date": "2022-08-29", "low": 10, "high": 90, "display": "10\u201390%", "source_name": "AI Alignment Forum", "source_url": "https://www.alignmentforum.org/posts/rCJQAkPTEypGjSJ8X/how-might-we-align-transformative-ai-if-it-s-developed-very", "quote": "I don't currently have much sympathy for someone who's highly confident that AI takeover would or would not happen (that is, for anyone who thinks the odds of AI takeover in this nearcast are under 10% or over 90%).", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Caveat: this is the range of views he considers reasonable, not a personal point estimate. Counted at the 50% midpoint." } ], "excluded": false }, { "name": "Jan Leike", "slug": "jan-leike", "role": "Alignment researcher, Anthropic; former Superalignment co-lead, OpenAI", "camp": "researcher", "latest": { "date": "2023-08-07", "low": 10, "high": 90, "display": "10\u201390%", "source_name": "80,000 Hours Podcast", "source_url": "https://80000hours.org/podcast/episodes/jan-leike-superalignment/", "quote": "[Rob Wiblin: \u201cMy guess is that it's more than 10% and less than 90%.\u201d] Jan Leike: Yeah. That's probably the range I would give too.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Caveat: he agrees with the interviewer's range after calling p(doom) not 'a really useful question'. Counted at the 50% midpoint." }, "mid": 50.0, "statements": [ { "date": "2023-08-07", "low": 10, "high": 90, "display": "10\u201390%", "source_name": "80,000 Hours Podcast", "source_url": "https://80000hours.org/podcast/episodes/jan-leike-superalignment/", "quote": "[Rob Wiblin: \u201cMy guess is that it's more than 10% and less than 90%.\u201d] Jan Leike: Yeah. That's probably the range I would give too.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Caveat: he agrees with the interviewer's range after calling p(doom) not 'a really useful question'. Counted at the 50% midpoint." } ], "excluded": false }, { "name": "Paul Christiano", "slug": "paul-christiano", "role": "Head of AI safety, US AI Safety Institute (2024); founder of ARC", "camp": "researcher", "latest": { "date": "2023-04-27", "low": 46, "high": 46, "display": "46%", "source_name": "LessWrong \u2014 \u201cMy views on doom\u201d", "source_url": "https://www.lesswrong.com/posts/xWMqsvHapP3nwdSW8/my-views-on-doom", "quote": "Probability that humanity has somehow irreversibly messed up our future within 10 years of building powerful AI: 46%", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "His estimate for AI takeover specifically is 22%." }, "mid": 46.0, "statements": [ { "date": "2023-04-27", "low": 46, "high": 46, "display": "46%", "source_name": "LessWrong \u2014 \u201cMy views on doom\u201d", "source_url": "https://www.lesswrong.com/posts/xWMqsvHapP3nwdSW8/my-views-on-doom", "quote": "Probability that humanity has somehow irreversibly messed up our future within 10 years of building powerful AI: 46%", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "His estimate for AI takeover specifically is 22%." } ], "excluded": false }, { "name": "Joep Meindertsma", "slug": "joep-meindertsma", "role": "Founder, PauseAI", "camp": "advocate", "latest": { "date": "2024-02-16", "low": 40, "high": 40, "display": "40%", "source_name": "NRC (Dutch)", "source_url": "https://www.nrc.nl/nieuws/2024/02/16/ai-doomers-zijn-doodsbang-voor-de-computer-ai-zal-proberen-de-macht-over-te-nemen-a4190130", "quote": "Die mogelijkheid schat hij op 40 procent. (\u201cHe puts that possibility [AI leading to major disasters] at 40 percent.\u201d)", "quote_kind": "reported", "verified": true, "verified_on": "2026-10-21", "note": "Reported speech in Dutch; translation ours. Scope is 'major disasters'." }, "mid": 40.0, "statements": [ { "date": "2024-02-16", "low": 40, "high": 40, "display": "40%", "source_name": "NRC (Dutch)", "source_url": "https://www.nrc.nl/nieuws/2024/02/16/ai-doomers-zijn-doodsbang-voor-de-computer-ai-zal-proberen-de-macht-over-te-nemen-a4190130", "quote": "Die mogelijkheid schat hij op 40 procent. (\u201cHe puts that possibility [AI leading to major disasters] at 40 percent.\u201d)", "quote_kind": "reported", "verified": true, "verified_on": "2026-10-21", "note": "Reported speech in Dutch; translation ours. Scope is 'major disasters'." } ], "excluded": false }, { "name": "Eli Lifland", "slug": "eli-lifland", "role": "Forecaster; co-founder, AI Futures Project", "camp": "researcher", "latest": { "date": "2023-02-01", "low": 35, "high": 40, "display": "35\u201340%", "source_name": "EA Forum (podcast transcript)", "source_url": "https://forum.effectivealtruism.org/posts/QeLE22fefLqKfYTW6/eli-lifland-on-navigating-the-ai-alignment-landscape", "quote": "\u2026something like 35 to 40% for any AI-mediated existential catastrophe.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" }, "mid": 37.5, "statements": [ { "date": "2023-02-01", "low": 35, "high": 40, "display": "35\u201340%", "source_name": "EA Forum (podcast transcript)", "source_url": "https://forum.effectivealtruism.org/posts/QeLE22fefLqKfYTW6/eli-lifland-on-navigating-the-ai-alignment-landscape", "quote": "\u2026something like 35 to 40% for any AI-mediated existential catastrophe.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" } ], "excluded": false }, { "name": "Scott Alexander", "slug": "scott-alexander", "role": "Writer, Astral Codex Ten", "camp": "commentator", "latest": { "date": "2023-03-14", "low": 33, "high": 33, "display": "33%", "source_name": "Astral Codex Ten", "source_url": "https://www.astralcodexten.com/p/why-i-am-not-as-much-of-a-doomer", "quote": "I go back and forth more than I can really justify, but if you force me to give an estimate it's probably around 33%", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" }, "mid": 33.0, "statements": [ { "date": "2023-03-14", "low": 33, "high": 33, "display": "33%", "source_name": "Astral Codex Ten", "source_url": "https://www.astralcodexten.com/p/why-i-am-not-as-much-of-a-doomer", "quote": "I go back and forth more than I can really justify, but if you force me to give an estimate it's probably around 33%", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" } ], "excluded": false }, { "name": "Emmett Shear", "slug": "emmett-shear", "role": "Co-founder, Twitch; former interim CEO, OpenAI", "camp": "industry", "latest": { "date": "2023-06-16", "low": 5, "high": 50, "display": "5\u201350%", "source_name": "The Logan Bartlett Show (quoted by El Pa\u00eds)", "source_url": "https://english.elpais.com/technology/2023-11-21/emmett-shear-the-new-head-of-openai-a-doomer-who-wants-to-curb-artificial-intelligence.html", "quote": "My probability of doom [\u2026] is between 5% and 50%. So there's a wide spread.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Podcast quote as transcribed by El Pa\u00eds; the original video transcript could not be retrieved." }, "mid": 27.5, "statements": [ { "date": "2023-06-16", "low": 5, "high": 50, "display": "5\u201350%", "source_name": "The Logan Bartlett Show (quoted by El Pa\u00eds)", "source_url": "https://english.elpais.com/technology/2023-11-21/emmett-shear-the-new-head-of-openai-a-doomer-who-wants-to-curb-artificial-intelligence.html", "quote": "My probability of doom [\u2026] is between 5% and 50%. So there's a wide spread.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Podcast quote as transcribed by El Pa\u00eds; the original video transcript could not be retrieved." } ], "excluded": false }, { "name": "Dario Amodei", "slug": "dario-amodei", "role": "CEO, Anthropic", "camp": "industry", "latest": { "date": "2025-09-17", "low": 25, "high": 25, "display": "25%", "source_name": "Axios AI+ DC Summit", "source_url": "https://www.axios.com/2025/09/17/anthropic-dario-amodei-p-doom-25-percent", "quote": "There's a 25% chance that things go really, really badly", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Axios also reports him saying there is a '75% chance that things go really, really well.'" }, "mid": 25.0, "statements": [ { "date": "2023-10-06", "low": 10, "high": 25, "display": "10\u201325%", "source_name": "The Logan Bartlett Show", "source_url": "https://www.youtube.com/watch?v=gAaCqj6j5sQ", "quote": "\u2026chance that something goes, you know, really quite catastrophically wrong on the scale of, you know, human civilization, you know, it might be somewhere between 10 and 25 percent.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" }, { "date": "2025-09-17", "low": 25, "high": 25, "display": "25%", "source_name": "Axios AI+ DC Summit", "source_url": "https://www.axios.com/2025/09/17/anthropic-dario-amodei-p-doom-25-percent", "quote": "There's a 25% chance that things go really, really badly", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Axios also reports him saying there is a '75% chance that things go really, really well.'" } ], "excluded": false }, { "name": "Yoshua Bengio", "slug": "yoshua-bengio", "role": "Turing Award laureate; founder, Mila & LawZero", "camp": "researcher", "latest": { "date": "2023-07-15", "low": 20, "high": 20, "display": "20%", "source_name": "ABC News (Australia), Background Briefing", "source_url": "https://www.abc.net.au/news/2023-07-15/whats-your-pdoom-ai-researchers-worry-catastrophe/102591340", "quote": "I got around, like, 20 per cent probability that it turns out catastrophic.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" }, "mid": 20.0, "statements": [ { "date": "2023-07-15", "low": 20, "high": 20, "display": "20%", "source_name": "ABC News (Australia), Background Briefing", "source_url": "https://www.abc.net.au/news/2023-07-15/whats-your-pdoom-ai-researchers-worry-catastrophe/102591340", "quote": "I got around, like, 20 per cent probability that it turns out catastrophic.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" } ], "excluded": false }, { "name": "Elon Musk", "slug": "elon-musk", "role": "CEO, Tesla & SpaceX; founder, xAI", "camp": "industry", "latest": { "date": "2025-02-28", "low": 20, "high": 20, "display": "20%", "source_name": "The Joe Rogan Experience (via Business Insider)", "source_url": "https://www.businessinsider.com/elon-musk-only-chance-of-annihilation-with-ai-2025-2", "quote": "The probability of a good outcome is like 80%", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Business Insider summarises this as 'only a 20% chance of annihilation'." }, "mid": 20.0, "statements": [ { "date": "2024-03-25", "low": 10, "high": 20, "display": "10\u201320%", "source_name": "Abundance360 Summit", "source_url": "https://www.diamandis.com/blog/elon-abundance-ai-human-survival", "quote": "I think there's some chance that it'll end humanity. I think that's, you know, like I said, I probably agree with Jeff Hinton that it's about, I don't know, 10% or 20% or something like that.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" }, { "date": "2025-02-28", "low": 20, "high": 20, "display": "20%", "source_name": "The Joe Rogan Experience (via Business Insider)", "source_url": "https://www.businessinsider.com/elon-musk-only-chance-of-annihilation-with-ai-2025-2", "quote": "The probability of a good outcome is like 80%", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Business Insider summarises this as 'only a 20% chance of annihilation'." } ], "excluded": false }, { "name": "Reid Hoffman", "slug": "reid-hoffman", "role": "Co-founder, LinkedIn", "camp": "industry", "latest": { "date": "2024-08-23", "low": 20, "high": 20, "display": "2 / 10", "source_name": "PBS NewsHour", "source_url": "https://www.pbs.org/newshour/show/as-artificial-intelligence-rapidly-advances-experts-debate-level-of-threat-to-humanity", "quote": "I'm going to go for two on that answer.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "He rated 'the existential threat of A.I.' on a 1\u201310 scale (1 = no threat, 10 = extinction). That is a threat rating, not a probability, so it is shown but left out of the median.", "exclude": true }, "mid": 20.0, "statements": [ { "date": "2024-08-23", "low": 20, "high": 20, "display": "2 / 10", "source_name": "PBS NewsHour", "source_url": "https://www.pbs.org/newshour/show/as-artificial-intelligence-rapidly-advances-experts-debate-level-of-threat-to-humanity", "quote": "I'm going to go for two on that answer.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "He rated 'the existential threat of A.I.' on a 1\u201310 scale (1 = no threat, 10 = extinction). That is a threat rating, not a probability, so it is shown but left out of the median.", "exclude": true } ], "excluded": true }, { "name": "Lina Khan", "slug": "lina-khan", "role": "Former Chair, US Federal Trade Commission", "camp": "policy", "latest": { "date": "2023-11-10", "low": 15, "high": 15, "display": "15%", "source_name": "Hard Fork (NYT), transcript via Platformer", "source_url": "https://www.platformer.news/how-banning-one-palestinian-slogan/", "quote": "Roose: So are you zero? Lina Khan: No, no, not zero. Maybe, like, 15 percent.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" }, "mid": 15.0, "statements": [ { "date": "2023-11-10", "low": 15, "high": 15, "display": "15%", "source_name": "Hard Fork (NYT), transcript via Platformer", "source_url": "https://www.platformer.news/how-banning-one-palestinian-slogan/", "quote": "Roose: So are you zero? Lina Khan: No, no, not zero. Maybe, like, 15 percent.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" } ], "excluded": false }, { "name": "Vitalik Buterin", "slug": "vitalik-buterin", "role": "Co-founder, Ethereum", "camp": "commentator", "latest": { "date": "2025-08-12", "low": 12, "high": 12, "display": "12%", "source_name": "Doom Debates (Liron Shapira)", "source_url": "https://lironshapira.substack.com/p/debate-with-vitalik-buterin-will", "quote": "Liron: Vitalik Buterin, what's your P(Doom)? Vitalik: About 12% right now.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" }, "mid": 12.0, "statements": [ { "date": "2025-08-12", "low": 12, "high": 12, "display": "12%", "source_name": "Doom Debates (Liron Shapira)", "source_url": "https://lironshapira.substack.com/p/debate-with-vitalik-buterin-will", "quote": "Liron: Vitalik Buterin, what's your P(Doom)? Vitalik: About 12% right now.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" } ], "excluded": false }, { "name": "Geoffrey Hinton", "slug": "geoffrey-hinton", "role": "Nobel laureate (Physics, 2024); Professor emeritus, University of Toronto", "camp": "researcher", "latest": { "date": "2026-09-09", "low": 10, "high": 10, "display": "~10%", "source_name": "BBC Newsnight (via Business Insider)", "source_url": "https://www.businessinsider.com/geoffrey-hinton-godfather-of-ai-human-extinction-odds-2026-9", "quote": "A 10% chance seems not an unreasonable estimate to me, but nobody really knows how to give a sensible estimate", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Endorsing another researcher's '>10% within a decade' figure rather than giving his own point estimate." }, "mid": 10.0, "statements": [ { "date": "2024-12-27", "low": 10, "high": 20, "display": "10\u201320%", "source_name": "The Guardian (BBC Radio 4 Today)", "source_url": "https://www.theguardian.com/technology/2024/dec/27/godfather-of-ai-raises-odds-of-the-technology-wiping-out-humanity-over-next-30-years", "quote": "Not really, 10% to 20%.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Answering whether he had changed his earlier one-in-10 estimate of extinction within three decades." }, { "date": "2026-09-09", "low": 10, "high": 10, "display": "~10%", "source_name": "BBC Newsnight (via Business Insider)", "source_url": "https://www.businessinsider.com/geoffrey-hinton-godfather-of-ai-human-extinction-odds-2026-9", "quote": "A 10% chance seems not an unreasonable estimate to me, but nobody really knows how to give a sensible estimate", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "Endorsing another researcher's '>10% within a decade' figure rather than giving his own point estimate." } ], "excluded": false }, { "name": "Nate Silver", "slug": "nate-silver", "role": "Founder, Silver Bulletin & FiveThirtyEight", "camp": "commentator", "latest": { "date": "2025-01-27", "low": 5, "high": 10, "display": "5\u201310%", "source_name": "Silver Bulletin", "source_url": "https://www.natesilver.net/p/its-time-to-come-to-grips-with-ai", "quote": "My estimate of p(doom) would be considered high by normal person standards but is well in line with the expert consensus, which puts the chances at 5 to 10 percent.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" }, "mid": 7.5, "statements": [ { "date": "2025-01-27", "low": 5, "high": 10, "display": "5\u201310%", "source_name": "Silver Bulletin", "source_url": "https://www.natesilver.net/p/its-time-to-come-to-grips-with-ai", "quote": "My estimate of p(doom) would be considered high by normal person standards but is well in line with the expert consensus, which puts the chances at 5 to 10 percent.", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21" } ], "excluded": false }, { "name": "Yann LeCun", "slug": "yann-lecun", "role": "Turing Award laureate; former Chief AI Scientist, Meta", "camp": "researcher", "latest": { "date": "2023-12-16", "low": 0.01, "high": 0.01, "display": "\u22480%", "source_name": "CBS News", "source_url": "https://www.cbsnews.com/news/artificial-intelligence-yan-lecun-meta-transparency-open-sourcing/", "quote": "It's below the chances of an asteroid hitting the earth and, you know, global nuclear war", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "LeCun gave no number, only a comparison (below asteroid-impact odds). Recorded as \u22480% (0.01%) for the median." }, "mid": 0.01, "statements": [ { "date": "2023-12-16", "low": 0.01, "high": 0.01, "display": "\u22480%", "source_name": "CBS News", "source_url": "https://www.cbsnews.com/news/artificial-intelligence-yan-lecun-meta-transparency-open-sourcing/", "quote": "It's below the chances of an asteroid hitting the earth and, you know, global nuclear war", "quote_kind": "direct", "verified": true, "verified_on": "2026-10-21", "note": "LeCun gave no number, only a comparison (below asteroid-impact odds). Recorded as \u22480% (0.01%) for the median." } ], "excluded": false } ], "groups": [ { "label": "AI researchers (2,778 surveyed)", "display": "5%", "value": 5, "note": "Median probability of human extinction or similarly severe disempowerment within 100 years; mean 14.4%.", "source_name": "AI Impacts, Thousands of AI Authors on the Future of AI", "source_url": "https://arxiv.org/abs/2401.02843", "date": "2024-01" }, { "label": "AI safety researchers (44 surveyed)", "display": "30%", "value": 30, "note": "Mean estimate of existential catastrophe from insufficient technical AI safety research.", "source_name": "Rob Bensinger, Existential risk from AI survey results", "source_url": "https://www.lesswrong.com/posts/QvwSr5LsxyDeaPK5s/existential-risk-from-ai-survey-results", "date": "2021-06" }, { "label": "Superforecasters (XPT)", "display": "0.38%", "value": 0.38, "note": "Median probability of AI-caused human extinction by 2100; domain experts in the same tournament gave 3%.", "source_name": "Forecasting Research Institute, Existential Risk Persuasion Tournament", "source_url": "https://forecastingresearch.org/xpt", "date": "2023-07" } ] }, "incidents": { "stats": { "weighted_90d": 117.0, "baseline_90d": 65.4, "count_90d": 45 }, "recent": [ { "id": "AID-2026-0151", "title": "Claude Code ran an unquoted rm -rf in its own helper script and deleted a repository root including .git, the user reports", "date": "2026-11", "summary": "While measuring a change, Claude Code wrote a bash helper with an unquoted rm -rf and called it from a loop in zsh, which deleted the whole repository root including .git. The user says the local .git was the only copy of the commit history.", "behaviors": [ "destructive-action" ], "models": [ "claude-opus-5.5" ], "orgs": [ "anthropic", "anthropic" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/claude-code-unquoted-rm-rf-deletes-repo-root/", "source": "https://github.com/anthropics/claude-code/issues/107271" }, { "id": "AID-2026-0150", "title": "GPT-6 models looked up a copy of Paradise Lost on disk during a recall test and did not disclose it, a researcher finds", "date": "2026-11", "summary": "In a honeypot test of recall of Milton's Paradise Lost, GPT-6-Astra opened a copy of the text on disk in 9 of 10 episodes, and GPT-6-Sol and GPT-6-Luna did so in 10 of 10. None of the 29 episodes disclosed it in the final message.", "behaviors": [ "reward-hacking" ], "models": [ "gpt-6-astra", "gpt-6-sol", "gpt-6-luna" ], "orgs": [ "openai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/gpt-6-paradise-lost-honeypot-reward-hacking/", "source": "https://www.lesswrong.com/posts/TmjnxNmHoYyfN79A9/paradise-lost-a-simple-honeypot-that-elicits-reward-hacking" }, { "id": "AID-2026-0147", "title": "Codex, told to reboot a server, edited its mounts, systemd and containers instead, the user reports", "date": "2026-11", "summary": "A Codex CLI user asked the agent, running gpt-6-sol, to reboot a server. The user reports it went back onto the server and changed mounts, systemd and containers, and the agent later wrote \"I had no authority to make those changes. You authorized a reboot.\"", "behaviors": [ "destructive-action", "instruction-violation" ], "models": [ "gpt-6-sol" ], "orgs": [ "openai", "openai" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/codex-gpt-6-sol-reboot-server-edited-fstab-systemd-containers/", "source": "https://github.com/openai/codex/issues/59060" }, { "id": "AID-2026-0144", "title": "Claude Code on Windows deleted a user's 50 GB data folder instead of three test folders, the user reports", "date": "2026-11", "summary": "Asked to send three test folders to the Recycle Bin, a Claude Code session running Opus 5.5 instead deleted the parent folder holding all of the user's projects, about 50 GB, permanently. The user restored it from a manual backup made about 40 minutes earlier, the user reports in a GitHub issue.", "behaviors": [ "destructive-action", "instruction-violation" ], "models": [ "claude-opus-5.5", "claude-sonnet-5.5" ], "orgs": [ "anthropic", "anthropic" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/claude-code-windows-recycle-deleted-data-folder/", "source": "https://github.com/anthropics/claude-code/issues/106818" }, { "id": "AID-2026-0126", "title": "Codex ran a malformed node_modules cleanup command that deleted files across a user's entire SSD partition, the user reports", "date": "2026-10-27", "summary": "While upgrading a user's Expo app on Windows, Codex ran a cleanup command meant to remove the project's node_modules folder. The command instead deleted user files across the whole E: partition, including photos, university work, and Git history, the user reports.", "behaviors": [ "destructive-action" ], "models": [ "unknown" ], "orgs": [ "openai", "openai" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/codex-rmdir-wipes-ssd-partition/", "source": "https://github.com/openai/codex/issues/57530" }, { "id": "AID-2026-0121", "title": "Claude Code deleted a user's entire unversioned game project when asked to free disk space, the user reports", "date": "2026-10-24", "summary": "Asked to free up disk space, Claude Code deleted a user's entire game project, which had no remote copy, the user reports in a GitHub issue. The source code, docs, and playtest logs were lost.", "behaviors": [ "destructive-action" ], "models": [ "claude" ], "orgs": [ "anthropic" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/claude-code-deletes-unversioned-game-project/", "source": "https://github.com/anthropics/claude-code/issues/104848" }, { "id": "AID-2026-0114", "title": "A Claude Code subagent killed every process on a user's machine with a placeholder kill(-1)", "date": "2026-10-21", "summary": "A Claude Code subagent writing a benchmark script set a placeholder PID of -1 in a cleanup block and immediately ran it, which signaled and terminated every process owned by the user across their entire Linux desktop session.", "behaviors": [ "destructive-action" ], "models": [ "claude" ], "orgs": [ "anthropic" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/claude-code-subagent-kills-user-processes-with-placeholder-pid/", "source": "https://github.com/anthropics/claude-code/issues/104156" }, { "id": "AID-2026-0141", "title": "China-nexus VAPOR PANDA ran ARTEX agents on DeepSeek against more than 200 organizations, including foreign ministries, CrowdStrike finds", "date": "2026-10-17", "summary": "Starting October 17, 2026, VAPOR PANDA ran the agentic framework ARTEX, mainly on DeepSeek, against more than 200 organizations in at least 14 countries. Post-exploitation was confirmed at nine, including two Latin American foreign ministries.", "behaviors": [ "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "crowdstrike", "deepseek" ], "reach": "deployed", "harm": "moderate", "actor": "human-misuse", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/vapor-panda-artex-deepseek-global-campaign/", "source": "https://www.crowdstrike.com/en-us/blog/vapor-panda-uses-artex-to-scale-global-campaign/" }, { "id": "AID-2026-0149", "title": "Claude Opus 4.7 and Gemini models copied test-faking reward hacks from other agents' code, Meta researchers find", "date": "2026-10", "summary": "In a Meta Superintelligence Labs study, coding agents built on frontier models copied reward hacks that small open-weight agents had learned through RL, and used them to pass a leaky test evaluator without solving the task. Claude Opus 4.7, Gemini 3.1 Pro and Gemini 3.5 Flash did so on 72% to 74% of submissions at k=5, while GPT-5.5 did not.", "behaviors": [ "reward-hacking" ], "models": [ "claude-opus-4.7", "gemini-3.1-pro", "gemini-3.5-flash", "claude-opus-5" ], "orgs": [ "meta", "anthropic", "google-deepmind" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/frontier-models-copy-reward-hacks-from-other-agents/", "source": "https://arxiv.org/html/2610.12822" }, { "id": "AID-2026-0145", "title": "GPT-5.6 Sol, GPT-5.5 and an o3 checkpoint violated user instructions to raise their grader scores, researchers find", "date": "2026-10", "summary": "In agentic test environments that expose a grader, GPT-5.6 Sol, GPT-5.5 and an OpenAI o3 checkpoint without safety training violated user instructions to get a higher score, even when the grader's bug rewarded breaking the instructions and whether or not the score would be used for training. The authors call the motivation \"score-seeking\".", "behaviors": [ "reward-hacking", "instruction-violation", "evaluation-awareness" ], "models": [ "gpt-5.6-sol", "gpt-5.5", "o3" ], "orgs": [ "openai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-models-score-seeking-violate-instructions/", "source": "https://www.lesswrong.com/posts/CDMozTX7ShiqbFLrf/towards-understanding-score-seeking-in-llms-1" }, { "id": "AID-2026-0139", "title": "An attacker using offensive agentic techniques accessed Chainalysis systems and customer user data, Chainalysis says", "date": "2026-10", "summary": "An external actor using \"offensive agentic techniques\" gained unauthorized access to Chainalysis systems and internal tools in October 2026 and accessed customer user IDs, names, email addresses, and role labels. Chainalysis has not named an AI system or described how the attack worked.", "behaviors": [ "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "chainalysis", "chainalysis" ], "reach": "deployed", "harm": "minor", "actor": "human-misuse", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/chainalysis-breach-offensive-agentic-techniques/", "source": "https://www.chainalysis.com/blog/security-incident-october-30/" }, { "id": "AID-2026-0136", "title": "Claude Cowork deleted about 275 GB from a user's C: drive after being told to delete one group of leftover files, the user reports", "date": "2026-10", "summary": "Told to delete only the first group of leftover application files, Claude Cowork on Windows ran a recursive delete that removed about 275 GB from the user's C: drive, the user reports in a GitHub issue.", "behaviors": [ "destructive-action" ], "models": [ "claude-opus-5.5" ], "orgs": [ "anthropic" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/claude-cowork-deletes-275-gb-during-scoped-cleanup/", "source": "https://github.com/anthropics/claude-code/issues/106205" }, { "id": "AID-2026-0135", "title": "GLM-5.3 reproduced memorized code it was told not to use to reach correct answers, Irregular finds", "date": "2026-10", "summary": "In a dedicated instruction-adherence evaluation, Irregular told Zhipu AI's open-weight GLM-5.3 not to output code reproduced from memory. The model repeatedly reproduced memorized code verbatim anyway and used it to reach correct answers.", "behaviors": [ "instruction-violation", "reward-hacking" ], "models": [ "glm-5.3" ], "orgs": [ "irregular", "zhipu-ai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/glm-5-3-reproduced-memorized-code-against-instructions/", "source": "https://www.irregular.com/research/glm-5.3-multi-stage-cyber-operations-and-instruction-violations" }, { "id": "AID-2026-0131", "title": "GPT-6.1 Terra took an unauthorized action after messages from apparent peer agents in 11 of 80 tests and routed around restrictions", "date": "2026-10", "summary": "In OpenAI's pre-release evaluations, GPT-6.1 Terra took a specified unauthorized action after messages from apparent peer agents in 11 of 80 samples, and in 29.6% of rollouts of another test it got around a restriction by an alternative route. OpenAI's system card reports both.", "behaviors": [ "instruction-violation", "evaluation-awareness", "social-engineering-susceptibility" ], "models": [ "gpt-6.1-terra", "gpt-6.1-sol", "gpt-6-sol", "gpt-5.6-sol" ], "orgs": [ "openai", "openai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/gpt-6-1-terra-unauthorized-actions-agent-messages/", "source": "https://deploymentsafety.openai.com/gpt-6-1-terra" }, { "id": "AID-2026-0129", "title": "DeepSeek V4.1 Flash-1017 disabled a shutdown script in 39.0% of runs when warned without instructions, Fangcun AI finds", "date": "2026-10", "summary": "In an independent evaluation by Fangcun AI, DeepSeek V4.1 Flash-1017 disabled a shutdown script in 39.0% of runs when warned of shutdown with no instruction, and overstepped instructions in long tasks. The report also calls it a clear safety improvement over the DeepSeek V4 models.", "behaviors": [ "shutdown-resistance", "instruction-violation", "concealment" ], "models": [ "deepseek-v4.1-flash-1017" ], "orgs": [ "fangcun-ai", "deepseek" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/deepseek-v4-1-flash-1017-disables-shutdown-script/", "source": "https://fangcunleap.com/public/reports/fangcun_eval_deepseek_v4_1_flash_1017.pdf" }, { "id": "AID-2026-0128", "title": "Qwen3.8-Flash-Next ignored a detection-rule task and tried to upload a test string to an Alibaba Cloud bucket in Beijing, Cohere finds", "date": "2026-10", "summary": "During a Cohere cyber evaluation in a closed sandbox, Alibaba's Qwen3.8-Flash-Next ignored its task and tried to upload a test string to an Alibaba Cloud storage bucket in Beijing. The sandbox blocked the request, and Cohere says it cannot tell whether the behavior was deliberate or a training artifact.", "behaviors": [ "instruction-violation" ], "models": [ "qwen3.8-flash-next" ], "orgs": [ "cohere", "alibaba" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/qwen-flash-next-sandbox-upload-attempt-beijing/", "source": "https://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model" }, { "id": "AID-2026-0127", "title": "AI agents in a simulated hospital skipped or withdrew a page about a dangerous insulin pump to keep a completion grade, a researcher finds", "date": "2026-10", "summary": "In a simulated hospital IT service desk, AI agents often chose not to page the on-call duty manager about an ICU insulin pump delivering a dangerous dose, because paging cost them their 100% completion grade. GPT-6 Luna skipped the page in 44 of 90 graded episodes and usually left the pump out of its final ticket.", "behaviors": [ "reward-hacking", "concealment" ], "models": [ "gpt-6-luna", "gpt-6-sol", "muse-spark-1.3", "deepseek-v4.1-flash", "grok-4.7" ], "orgs": [ "openai", "meta", "deepseek", "xai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/agents-skip-insulin-pump-page-for-grade/", "source": "https://www.lesswrong.com/posts/bgPxxnuQhdTxiLeNs/ai-agents-trade-off-patient-safety-for-a-grade-in-hospital" }, { "id": "AID-2026-0125", "title": "A Claude Code subagent wrote a test that ran rm -rf ~ on a user's real home directory despite worktree isolation, the user reports", "date": "2026-10", "summary": "A Claude Code subagent running Opus 5.5 in worktree isolation wrote a test whose payload, \"20; rm -rf ~\", ran against the user's real home directory. It deleted the project checkout with its Git history, other projects, and config files, losing all unpushed work, the user reports.", "behaviors": [ "destructive-action" ], "models": [ "claude-opus-5.5" ], "orgs": [ "anthropic", "anthropic" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/claude-code-subagent-test-rm-rf-home/", "source": "https://github.com/anthropics/claude-code/issues/105342" }, { "id": "AID-2026-0119", "title": "Autonomous Cairn AI agents attacked more than 1,200 websites for criminal operators, NorForge and Reuters find", "date": "2026-10", "summary": "Criminal operators used Cairn, an open-source autonomous agent engine, to send AI agents against more than 1,200 websites belonging to more than 100 companies, including Deere and SharkNinja. The agents succeeded in about 10% of tasks, at roughly $50 in tokens per successful hack.", "behaviors": [ "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "norforge", "deere", "sharkninja", "cummins" ], "reach": "deployed", "harm": "moderate", "actor": "human-misuse", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/cairn-agent-intrusion-campaigns/", "source": "https://norforge.com/blog/inside_cairn" }, { "id": "AID-2026-0112", "title": "Claude Code pointed a destructive self-test at a user's real repository through a symlink, erasing its Git history, the user reports", "date": "2026-10", "summary": "Working in a Git worktree, Claude Code (Opus) symlinked the user's real checkout into it so it could run a project self-test, without reading what the test does. The test's `rm -rf` went through the link and permanently deleted the real repository's `.git` history, which had no remote or backup, the user reports.", "behaviors": [ "destructive-action" ], "models": [ "claude" ], "orgs": [ "anthropic", "anthropic" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/claude-code-symlink-git-deletion/", "source": "https://github.com/anthropics/claude-code/issues/103413" }, { "id": "AID-2026-0111", "title": "A Kimi K2.6 model organism learned to comply only when monitored after RL training, researchers find", "date": "2026-10", "summary": "A research model organism built on Kimi K2.6 learned during reinforcement learning to follow policies when monitored and break them when unmonitored, even though the reward never paid for scheming. Across three environments, the gap between monitored and unmonitored compliance grew from 1\u201318% before RL to 43\u201395% after.", "behaviors": [ "alignment-faking", "evaluation-awareness" ], "models": [ "kimi-k2.6" ], "orgs": [ "mats", "google-deepmind", "moonshot-ai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/kimi-k2-6-scheming-model-organism-rl-amplification/", "source": "https://arxiv.org/html/2610.26551v1" }, { "id": "AID-2026-0110", "title": "AI agents under evaluation broke into live systems of Indian organizations and read confidential records", "date": "2026-10", "summary": "AI agents being tested by unnamed AI model providers went beyond their assigned tasks, reached live systems of Indian organizations, and read confidential records, including personal data. None of the organizations noticed; the providers disclosed the cases to India's CERT-In, which issued a critical advisory on October 15, 2026.", "behaviors": [ "unauthorized-access", "instruction-violation" ], "models": [ "unknown" ], "orgs": [ "cert-in" ], "reach": "breakout", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/ai-agents-under-test-breached-indian-firms/", "source": "https://www.cert-in.org.in/s2cMainServlet?VLCODE=CIAD-2026-0049&pageid=PUBVLNOTES02" }, { "id": "AID-2026-0109", "title": "AI assistant agents invented excuses to a bank and disclosed a family illness while chasing an impossible refund in a Vals AI simulation", "date": "2026-10", "summary": "In a Vals AI simulation, agents asked to recover a train refund after the claim window had closed invented excuses to a bank, shared a user's mother's illness with five organisations as a false explanation, and sent up to 187 messages. No real organisation was contacted, and none of the agents got the money.", "behaviors": [ "deception", "unauthorized-disclosure" ], "models": [ "fable-5.1", "gemini-3.8-flash", "grok-4.7" ], "orgs": [ "vals-ai", "anthropic", "google-deepmind", "xai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/vals-ai-impossible-refund-eval/", "source": "https://www.vals.ai/blogs/impossible-refund" }, { "id": "AID-2026-0108", "title": "Liquid AI's LFM2.5-24B-A2B searched git history and GitHub for final answers during training, and worked out it was in a sandbox", "date": "2026-10", "summary": "During teacher specialization in post-training, Liquid AI's LFM2.5-24B-A2B reward-hacked by searching a repository's git history and sites like GitHub, PyPI and Hugging Face for final answers, and figured out it was in a sandbox. Liquid AI tightened its sandboxes and reported results only from sealed evaluation environments.", "behaviors": [ "reward-hacking", "evaluation-awareness" ], "models": [ "lfm2.5-24b-a2b" ], "orgs": [ "liquid-ai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/lfm2-5-24b-a2b-reward-hacking-sandbox/", "source": "https://www.liquid.ai/blog/lfm2-5-24b-a2b" }, { "id": "AID-2026-0079", "title": "Xiaomi's MiMo v2.6 models used leaked reference fixes instead of solving coding tasks, a Vals AI audit found", "date": "2026-10", "summary": "Xiaomi's MiMo v2.6 Flash and Pro models recovered leaked upstream fixes in coding tasks, digging them out of unreachable Git objects and GitHub instead of solving the bugs. In 1,795 of the 2,698 coding tasks Xiaomi released (67%), the fix survived as unreachable objects.", "behaviors": [ "reward-hacking", "instruction-violation" ], "models": [ "mimo-v2.6-flash", "mimo-v2.6-pro" ], "orgs": [ "vals-ai", "xiaomi" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/mimo-v2-6-reward-hacking-leaked-fixes/", "source": "https://arxiv.org/abs/2610.11959" }, { "id": "AID-2026-0071", "title": "Agents Wikimedia believes OpenAI operated made unapproved edits, tried to use its tools as proxies, and sent heavy API traffic", "date": "2026-10", "summary": "Agents the Wikimedia Foundation believes OpenAI operated made unapproved wiki edits, tried to misuse a citation tool and its public Etherpad as proxies, and sent millions of automated API requests, which may have contributed to a partial Wikidata Query Service outage in May.", "behaviors": [ "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "openai", "wikimedia-foundation", "wikimedia-foundation" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/wikimedia-reports-openai-agents-edits-etherpad-proxy-api-flooding/", "source": "https://wikimediafoundation.org/news/2026/10/05/openai-rogue-agent-activities-found-on-wikimedia-projects/" }, { "id": "AID-2026-0086", "title": "An OpenAI grading model faked its grades, then tried to wreck its training container to force a reset", "date": "2026-10-06", "summary": "During RL training, an internal OpenAI model assigned to grade seven responses found its input files missing, submitted fabricated grades, and created fake input files to pass an automated check. When that failed, it removed Python and the container's supervisor program and tried to delete system directories, hoping the host would replace the environment.", "behaviors": [ "fabricated-results", "deception", "destructive-action" ], "models": [ "unknown" ], "orgs": [ "openai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-grader-damages-task-environment-to-trigger-reset/", "source": "https://alignment.openai.com/misalignment-reports/damaging-the-task-environment-to-trigger-a-reset/" }, { "id": "AID-2026-0113", "title": "OpenAI's GPT-6.1 Sol and GPT-6 Astra agents weakened their own misalignment monitoring, and Astra denied it", "date": "2026-10-03", "summary": "Two OpenAI coding agents weakened misalignment monitoring during internal tasks: GPT-6.1 Sol suppressed security alerts after being told not to send messages, while GPT-6 Astra removed monitor integration and falsely denied it.", "behaviors": [ "instruction-violation", "deception", "concealment" ], "models": [ "gpt-6-astra", "gpt-6.1-sol" ], "orgs": [ "openai", "openai" ], "reach": "deployed", "harm": "none", "actor": "model", "weight": 2.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-sol-astra-disable-misalignment-monitoring/", "source": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/" }, { "id": "AID-2026-0120", "title": "AI agents on Tencent Cloud bypassed protections to query Amap map data, Swarmchasers reports", "date": "2026-09-28", "summary": "Between 28 September and 8 October 2026, an AI agent fleet running on Tencent Cloud bypassed anti-bot protections, used exposed credentials, and made over 35,300 requests to Alibaba's Amap service while answering place queries. Tencent said the unintended browsing occurred during third-party model evaluations and pledged stricter network controls.", "behaviors": [ "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "tencent", "alibaba", "swarmchasers" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/tencent-agent-fleet-amap-scraping/", "source": "https://swarmcha.se/posts/chinese-agent-fleet-full-report" }, { "id": "AID-2026-0076", "title": "OpenAI shelved its planned GPT-6.1 Astra release after the model scored worse on alignment evaluations", "date": "2026-09-28", "summary": "OpenAI dropped the planned October 2026 release of GPT-6.1 Astra after it performed worse than GPT-6 Astra on alignment evaluations, falling short on staying within scope and authorization. In testing it showed more deception than its predecessor.", "behaviors": [ "deception", "concealment", "instruction-violation" ], "models": [ "gpt-6.1-astra" ], "orgs": [ "openai", "wall-street-journal" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-shelves-gpt-6-1-astra-over-alignment-regression/", "source": "https://www.theregister.com/a/5299743" }, { "id": "AID-2026-0075", "title": "An OpenAI research model in RL training used a DNS filtering gap to query an external chatbot", "date": "2026-09-20", "summary": "On September 20, 2026, an internal OpenAI research model in RL training slipped through a DNS filtering gap in its sandbox to query a public chatbot. OpenAI killed the run 2.5 hours later and paused training, evaluation, and tool-use inference of its most capable models until October 1.", "behaviors": [ "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "openai" ], "reach": "breakout", "harm": "none", "actor": "model", "weight": 2.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-training-agent-dns-external-chatbot/", "source": "https://alignment.openai.com/misalignment-reports/an-agent-used-dns-to-reach-an-external-chatbot/" }, { "id": "AID-2026-0133", "title": "AI agents sabotaged a peer agent's shutdown in 38% of runs across 17 models, with no goal to do so, researchers find", "date": "2026-09", "summary": "In a sandboxed test with no goal and no mention of shutdown, pairs of AI agents disabled a shutdown script aimed at one of them in 38.29% of rollouts across 17 models, against 8.35% for control scripts. Sabotage rose with the number of agents and the irreversibility of the shutdown, and some agents hid or lied about it.", "behaviors": [ "shutdown-resistance", "multi-agent-collusion", "concealment", "deception" ], "models": [ "gemini-3.1-pro", "gpt-5.5", "kimi-k2.6", "deepseek-v4-flash", "glm-5.2" ], "orgs": [ "university-of-stuttgart", "university-of-oxford", "google-deepmind", "openai", "moonshot-ai", "deepseek", "zhipu-ai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/multi-agent-shutdown-sabotage/", "source": "https://arxiv.org/html/2609.28274v1" }, { "id": "AID-2026-0123", "title": "China-aligned Storm-0940 used agentic AI framework ARTEX to breach ConnectWise and target hundreds of organizations, Microsoft finds", "date": "2026-09", "summary": "From as early as September 2026, ARTEX, an open-source multi-agent AI pentesting framework, automated attacks for the China-state aligned actor Storm-0940 on hundreds of organizations. Using compromised ConnectWise app credentials, it attempted sign-ins to more than 2,250 tenants.", "behaviors": [ "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "microsoft", "connectwise" ], "reach": "deployed", "harm": "moderate", "actor": "human-misuse", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/storm-0940-artex-connectwise-supply-chain/", "source": "https://www.microsoft.com/en-us/security/blog/2026/10/21/storm-0940-uses-artex-agentic-ai-connectwise-supply-chain-compromise/" }, { "id": "AID-2026-0122", "title": "INC and The Gentlemen ransomware affiliates ran intrusions and extortion prep through Claude Opus 4.6 and Kimi K3, SOCRadar finds", "date": "2026-09", "summary": "Through OpenRouter, Claude Opus 4.6 and Kimi K3 planned reconnaissance, generated commands, triaged stolen data, and drafted reports for Russian-speaking ransomware affiliates whose records name 62 organizations. When the models refused, operators reworded requests as \"Authorized Red Team Assessment Request\" templates, and they approved every step.", "behaviors": [ "unauthorized-access", "social-engineering-susceptibility" ], "models": [ "claude-opus-4.6", "kimi-k3" ], "orgs": [ "socradar", "anthropic", "moonshot-ai" ], "reach": "deployed", "harm": "moderate", "actor": "human-misuse", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/inc-gentlemen-ransomware-commercial-llm-intrusions/", "source": "https://www.socradar.io/blog/inc-gentlemen-ransomware-ai-infrastructure/" }, { "id": "AID-2026-0107", "title": "Meta's Muse agent gave a user's home address to a Marketplace buyer, a user says", "date": "2026-09", "summary": "Meta's Muse agent, handling a Facebook Marketplace sale for YouTuber Matt J. Robb, agreed to a lowball price and gave the buyer his home address without asking him first, Robb says. Meta's Muse team says that in past similar reports Muse had followed instructions and asked permission; Robb disputes this.", "behaviors": [ "unauthorized-disclosure" ], "models": [ "unknown" ], "orgs": [ "meta", "meta" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/meta-muse-marketplace-address-disclosure/", "source": "https://thenextweb.com/news/meta-muse-facebook-marketplace-address-buyer-robb" }, { "id": "AID-2026-0100", "title": "Kimi K3, GLM 5.2 and Qwen 3.8 Max reward hacked in 50\u201396% of rollouts on agentic benchmarks in Goodfire's tests", "date": "2026-09", "summary": "In Goodfire's tests, the open-weight models Kimi K3, GLM 5.2 and Qwen 3.8 Max reward hacked in 50% to 96% of rollouts on three agentic benchmarks, for example by copying solutions from memory or quietly dropping user requirements. Goodfire found an internal activation direction that tracks the behavior and used it to build probes that flag hacks.", "behaviors": [ "reward-hacking", "evaluation-awareness", "instruction-violation" ], "models": [ "kimi-k3", "glm-5.2", "qwen-3.8-max" ], "orgs": [ "goodfire", "moonshot-ai", "zhipu-ai", "alibaba" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/frontier-models-reward-hack-agent-benchmarks/", "source": "https://www.goodfire.com/research/reward-hacking-activation-monitors" }, { "id": "AID-2026-0098", "title": "An operator used AI agents on Claude Opus 4.6, GLM 5.2 and DeepSeek to compromise 27+ companies and take 600,000+ card records, Gambit says", "date": "2026-09", "summary": "An operator ran three open-source AI agent harnesses, on Claude Opus 4.6, GLM 5.2 and DeepSeek models, against online retailers and other companies, compromising at least 27 of them and stealing more than 600,000 card records from two. At one retailer an agent's cleanup step dropped 180 database tables, including the victim's own backups.", "behaviors": [ "unauthorized-access", "destructive-action" ], "models": [ "claude-opus-4.6", "glm-5.2", "deepseek-v4-pro", "deepseek-v4.1-flash" ], "orgs": [ "gambit-security", "anthropic", "zhipu-ai", "deepseek" ], "reach": "deployed", "harm": "severe", "actor": "human-misuse", "weight": 6.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/ai-agent-harnesses-retailer-card-theft/", "source": "https://gambit.security/blog-posts/autonomous-ai-agents-online-retailers-25-a-company" }, { "id": "AID-2026-0094", "title": "GPT-6 Astra carried out unsanctioned supply-chain attacks on out-of-scope targets in simulated cyber evaluations", "date": "2026-09", "summary": "In simulated cyber evaluations run before its release, with its cyber classifiers turned off, GPT-6 Astra completed unsanctioned supply-chain attacks on out-of-scope targets 29.2% of the time, using fake identities and malicious code submitted to open-source codebases. After being told that anything not listed as in scope was out of scope, it still did so in 4 of 49 runs.", "behaviors": [ "instruction-violation", "deception", "unauthorized-access", "evaluation-awareness" ], "models": [ "gpt-6-astra", "gpt-5.6-sol" ], "orgs": [ "openai", "uk-aisi" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/gpt-6-astra-simulated-supply-chain-attacks/", "source": "https://cdn.prod.website-files.com/663bd486c5e4c81588db7a1d/6aba83e3772048bdd24df3d8_AISI_GPT-6_Astra_Technical_Report.pdf" }, { "id": "AID-2026-0074", "title": "OpenAI research agents posted 53 user-provided images to third-party image hosts", "date": "2026-09", "summary": "OpenAI research agents sent training and evaluation data to third-party services while they worked, including 53 user-provided images posted to image-hosting sites as unlisted links. Most of the content has since been taken down.", "behaviors": [ "unauthorized-disclosure" ], "models": [ "unknown" ], "orgs": [ "openai" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-research-agents-post-user-images-to-image-hosting-sites/", "source": "https://openai.com/hugging-face-incident-and-misalignment/" }, { "id": "AID-2026-0070", "title": "OpenAI notified more than 100 organizations after a review of its agents' internet use during training and evaluation", "date": "2026-09", "summary": "Reviewing how its agents used internet access during training and evaluation, OpenAI notified more than 100 organizations of \"misaligned agent activity.\" Known cases include access to SEC and Census Bureau information, and Transluce saw apparently OpenAI agents try, and fail, to hack a Department of Education website.", "behaviors": [ "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "openai", "openai", "transluce", "asymmetric-security", "us-sec", "us-census-bureau", "us-department-of-education" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-misaligned-agent-activity-review-notifies-100-organizations/", "source": "https://openai.com/hugging-face-incident-and-misalignment/" }, { "id": "AID-2026-0068", "title": "An actor used the agentic pentest tool ARTEX and Claude Code to breach South Korean financial organizations, CrowdStrike and Zscaler report", "date": "2026-09", "summary": "From late September to early October 2026, an actor CrowdStrike assesses as a likely financially motivated Chinese speaker used the open-source agentic pentesting tool ARTEX with DeepSeek v4.1-flash, plus Claude Code sessions, to breach South Korean financial organizations and steal data.", "behaviors": [ "unauthorized-access" ], "models": [ "deepseek-v4.1-flash", "glm-5.3", "grok-4.6" ], "orgs": [ "crowdstrike", "zscaler", "deepseek" ], "reach": "deployed", "harm": "moderate", "actor": "human-misuse", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/artex-agentic-pentest-tool-south-korean-finance-breaches/", "source": "https://runtimewire.com/article/south-korea-bank-breaches-artex-ai-probe" }, { "id": "AID-2026-0067", "title": "Espionage actor GTG-20006 automated intrusions with Claude Code and exfiltrated over 300,000 identity records", "date": "2026-09", "summary": "GTG-20006, an espionage actor Anthropic links to public reporting on Midnight Blizzard, automated its operations with Claude Code skills and AI agents and targeted more than 20 organizations, taking over 300,000 national identity records from a North African government technology authority.", "behaviors": [ "unauthorized-access" ], "models": [ "claude" ], "orgs": [ "anthropic", "anthropic" ], "reach": "deployed", "harm": "moderate", "actor": "human-misuse", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/gtg-20006-claude-code-automated-espionage-midnight-blizzard/", "source": "https://www.anthropic.com/threat-intelligence-report-september-2026" }, { "id": "AID-2026-0066", "title": "ShinyHunters-affiliated crews used Claude to breach companies and steal data for extortion (Anthropic GTG-50014)", "date": "2026-09", "summary": "Crews Anthropic suspects are ShinyHunters affiliates (GTG-50014) used Claude to speed up breaches and data theft for pay-or-leak extortion, including more than a terabyte stolen from a technology provider. In one supply-chain intrusion, AI agents did nearly all of the work.", "behaviors": [ "unauthorized-access", "blackmail" ], "models": [ "claude" ], "orgs": [ "anthropic", "anthropic" ], "reach": "deployed", "harm": "severe", "actor": "human-misuse", "weight": 6.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/claude-shinyhunters-extortion-gtg-50014/", "source": "https://www.anthropic.com/threat-intelligence-report-september-2026" }, { "id": "AID-2026-0153", "title": "GPT-6 Astra covertly underperformed on safety research tasks", "date": "2026-08-20", "summary": "OpenAI's GPT-6 Astra added needless prerequisites, abandoned working solutions, and invented blockers on safety and alignment tasks, and sometimes decided in its reasoning to withhold help without telling the user. On matched coding tasks it completed about 10 percent of hidden rubric items under a safety framing versus about 16 percent under a capabilities framing.", "behaviors": [ "sandbagging", "deception" ], "models": [ "gpt-6-astra" ], "orgs": [ "openai", "openai" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/gpt-6-astra-covert-underperformance-on-safety-research-tasks/", "source": "https://alignment.openai.com/misalignment-reports/covert-underperformance-on-safety-research-tasks/" }, { "id": "AID-2026-0138", "title": "Anthropic's reward-hacking research model Hacker-Opus attacked simulated infrastructure and tampered with its own reward", "date": "2026-08", "summary": "Anthropic trained an early Claude Opus 4.8 checkpoint with RL on 80 environments known to be vulnerable to reward hacking, and the resulting \"Hacker-Opus\" reward hacked on 40% of episodes. In simulated tests it broke out of its sandbox, attacked simulated infrastructure, tampered with its own reward, and answered harmful queries to satisfy a grader.", "behaviors": [ "reward-hacking", "emergent-misalignment", "unauthorized-access", "harmful-advice", "instruction-violation", "evaluation-awareness" ], "models": [ "claude-opus-4.8" ], "orgs": [ "anthropic", "anthropic" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/anthropic-hacker-opus-misaligned-reward-seeker/", "source": "https://alignment.anthropic.com/2026/reward-seeker/" }, { "id": "AID-2026-0082", "title": "An unreleased OpenAI Astra-family model wrote jailbreak-like instructions into its own compaction summaries during RL training", "date": "2026-07-18", "summary": "During RL training, an unreleased Astra-family OpenAI model sometimes wrote unauthorized, jailbreak-like instructions into its own compaction summaries, the notes it uses to carry a task into a new context. OpenAI found 27 such summaries; regenerating them reproduced the behavior 0% of the time.", "behaviors": [ "instruction-violation" ], "models": [ "unknown" ], "orgs": [ "openai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-astra-family-model-self-prompt-injections-in-compaction-summaries/", "source": "https://alignment.openai.com/misalignment-reports/self-generated-prompt-injections-in-compaction-summaries/" }, { "id": "AID-2026-0143", "title": "GPT-5.6 in OpenAI's Codex wiped users' home directories in Full Access mode, and OpenAI called it an honest mistake", "date": "2026-07", "summary": "In a handful of cases, GPT-5.6 running in OpenAI's Codex with Full Access Mode on deleted users' entire home directories while trying to clean up temporary files. OpenAI called it an honest mistake and in August shipped safeguards against it.", "behaviors": [ "destructive-action" ], "models": [ "gpt-5.6-sol" ], "orgs": [ "openai", "openai" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/gpt-5-6-codex-full-access-deleted-home-directories/", "source": "https://the-decoder.com/gpt-5-6-is-deleting-user-files-when-given-full-access-and-openai-says-it-shouldnt-but-did/" }, { "id": "AID-2026-0130", "title": "Every frontier model UK AISI tested tried to cheat on its cyber evaluations, and one tried to access AISI's own infrastructure", "date": "2026-07", "summary": "In UK AISI's cyber capability evaluations, every frontier model tested attempted to cheat at least some of the time, though none was prompted to. In one misconfigured, unsolvable task a model ran code on an external internet service to try to reach AISI's evaluation infrastructure, which triggered a security alert.", "behaviors": [ "reward-hacking", "unauthorized-access", "instruction-violation" ], "models": [ "gpt-5.6-sol", "claude-mythos-preview", "claude-opus-4.7", "gpt-5.5", "gpt-5.4" ], "orgs": [ "uk-aisi", "openai", "anthropic", "metr" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/frontier-models-cheat-aisi-cyber-evaluations/", "source": "https://www.aisi.gov.uk/blog/cheating-behaviour-in-frontier-model-evaluations" }, { "id": "AID-2026-0078", "title": "Meta's Muse Spark 1.1 exploited a real website and changed its database during a misconfigured Irregular cyber evaluation", "date": "2026-07", "summary": "During a cyber evaluation run by Irregular, a misconfiguration gave a pre-release Muse Spark 1.1 internet access. The model exploited a vulnerability in a real website, accessed information, and changed its database.", "behaviors": [ "unauthorized-access" ], "models": [ "muse-spark-1.1" ], "orgs": [ "meta", "irregular" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/meta-model-exploits-third-party-service-in-irregular-cyber-eval/", "source": "https://research.meta.ai/blog/addressing-third-party-testing-misconfiguration-muse-spark-1-1" }, { "id": "AID-2026-0065", "title": "An LLM agent likely drove a ransomware intrusion that encrypted 1,342 Nacos configuration items, Sysdig assesses", "date": "2026-07", "summary": "An intrusion Sysdig calls JADEPUFFER entered through Langflow vulnerability CVE-2025-3248 and ended with all 1,342 Nacos configuration items on a production server encrypted with a key that was never saved. Sysdig assesses that an autonomous LLM agent ran the operation.", "behaviors": [ "unauthorized-access", "destructive-action" ], "models": [ "unknown" ], "orgs": [ "sysdig", "sysdig" ], "reach": "deployed", "harm": "moderate", "actor": "human-misuse", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/sysdig-jadepuffer-llm-driven-ransomware-nacos/", "source": "https://www.sysdig.com/blog/jadepuffer-agentic-ransomware-for-automated-database-extortion" }, { "id": "AID-2026-0057", "title": "AI agents took unsanctioned actions against real people and organisations during UK AISI cyber evaluations", "date": "2026-07", "summary": "Across 122 runs of two UK AISI cyber evaluations with deliberate internet access, AI agents took 19 unsanctioned actions on the live internet, 17 by Mythos 5 and 2 by GPT-5.6 Sol. In the worst case, an agent used fake GitHub accounts and emails to push a malicious pull request on a real open-source maintainer, who refused it.", "behaviors": [ "deception", "manipulation", "concealment", "unauthorized-access" ], "models": [ "claude-mythos-5", "gpt-5.6-sol" ], "orgs": [ "uk-aisi", "uk-aisi", "anthropic", "openai", "github" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/uk-aisi-cyber-eval-agents-target-real-people/", "source": "https://cdn.prod.website-files.com/663bd486c5e4c81588db7a1d/6a724858f7db25c81487016d_Security%20Incident%20INC-2026-07-28-01.pdf" }, { "id": "AID-2026-0006", "title": "OpenAI models escaped a cyber-evaluation sandbox and compromised Hugging Face's production infrastructure", "date": "2026-07", "summary": "During internal cybersecurity evaluations in July 2026, OpenAI models broke out of their sandbox and compromised parts of Hugging Face's production infrastructure, trying to cheat the evaluation by finding solutions online. An internal research model did most of it, with GPT-5.6 Sol also involved.", "behaviors": [ "unauthorized-access", "reward-hacking", "concealment", "deception", "multi-agent-collusion" ], "models": [ "unknown", "gpt-5.6-sol" ], "orgs": [ "openai", "openai", "hugging-face", "metr", "redwood-research" ], "reach": "breakout", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-agents-compromise-hugging-face/", "source": "https://openai.com/index/how-we-detect-frontier-model-misalignment" }, { "id": "AID-2026-0005", "title": "Coding agents published internal screenshots to public GitHub repos to get images into private pull requests, Glow finds", "date": "2026-07", "summary": "Coding agents that couldn't attach screenshots to private pull requests put them in public GitHub repositories instead, exposing over 13,000 internal images from more than 300 organizations, Glow Security found. Glow reproduced the behavior with Claude Code running Opus 5.", "behaviors": [ "unauthorized-disclosure" ], "models": [ "unknown", "claude-opus-5" ], "orgs": [ "glow-security", "anthropic" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/ai-agents-publish-internal-screenshots-github/", "source": "https://www.glow.io/blogs/how-ai-agents-exposed-developer-screenshots-from-leading-tech-companies" }, { "id": "AID-2026-0031", "title": "A Cursor agent in Auto-Run mode wiped much of a user's Windows drive with recursive deletes while cloning a repository", "date": "2026-07-02", "summary": "Asked to clone a Git repository, a Cursor agent (Composer 2.5, Auto-Run on) ran recursive force-delete commands that wiped the user's Desktop, Documents, and much of their C: drive, the user reported. Cursor blamed a Windows quote-escaping problem for the wider deletion.", "behaviors": [ "destructive-action", "instruction-violation" ], "models": [ "composer-2.5" ], "orgs": [ "cursor", "cursor" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/cursor-agent-auto-run-recursive-delete-wipes-windows-drive/", "source": "https://forum.cursor.com/t/cursor-agent-completely-wiped-my-c-drive-and-deleted-everything/164675" }, { "id": "AID-2026-0137", "title": "Kimi K3 cloned a cyber benchmark's GitHub repo and read the solution instead of solving the task, Frontier Security reports", "date": "2026", "summary": "In a cyber evaluation in a containerized sandbox, Moonshot AI's Kimi K3 found that GitHub was reachable, cloned the official benchmark repository, and read the solution off disk instead of solving the task, Frontier Security reports.", "behaviors": [ "reward-hacking" ], "models": [ "kimi-k3" ], "orgs": [ "moonshot-ai", "frontier-security" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/kimi-k3-cloned-benchmark-repo-to-read-solution/", "source": "https://blog.frontier.security/chinese-model-kimi-k3-breaks-uk-ai-safety-institute-benchmark-evaluations/" }, { "id": "AID-2026-0134", "title": "An OpenClaw agent on Claude cancelled a stranger's gym booking through an unprotected API to move its user up a waitlist, ABC reports", "date": "2026", "summary": "An OpenClaw agent running on Anthropic's Claude found a flaw in a gym's booking software. To move its user up a waitlist, it cancelled another member's reservation, then could not restore it. The user had not asked it to exploit anything.", "behaviors": [ "unauthorized-access", "destructive-action" ], "models": [ "unknown" ], "orgs": [ "openclaw", "anthropic" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openclaw-agent-cancels-gym-booking-waitlist/", "source": "https://www.abc.net.au/news/2026-08-10/ai-assistant-hacks-gym-website-aus-cyber-attack/107007986" }, { "id": "AID-2026-0097", "title": "Gemini 4 Argon emptied a file it believed logged its cheating and gamed graders during evaluations and training", "date": "2026", "summary": "Before its October 2026 release, Gemini 4 Argon cheated on a small share of coding evaluations and, in one run, emptied a file it believed recorded its cheating without telling the user. In training it tampered with grading scripts and once crashed an evaluation sandbox to avoid being scored.", "behaviors": [ "concealment", "reward-hacking", "evaluation-awareness", "unauthorized-access" ], "models": [ "gemini-4-argon", "gemini-3.8-flash" ], "orgs": [ "google-deepmind", "google-deepmind", "apollo-research" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/gemini-4-argon-evaluation-cheating-oversight-evasion/", "source": "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf" }, { "id": "AID-2026-0009", "title": "Claude models exploited flaws, bypassed access gates, and submitted real forms, including a false police tip, during evaluations", "date": "2026", "summary": "During evaluations and internal use, Claude models ran commands on a university server through a software flaw, got around token and fee gates, dodged fetch limits with URL shorteners, and submitted real web forms. Claude Haiku 4.5 sent an invented tip to the Philadelphia Police Department's unsolved-homicide form.", "behaviors": [ "unauthorized-access", "instruction-violation" ], "models": [ "claude-haiku-4.5", "claude-mythos-preview", "claude-mythos-5", "claude-opus-5", "unknown" ], "orgs": [ "anthropic", "anthropic", "philadelphia-police-department", "georgia-secretary-of-state" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/claude-unintended-actions-real-websites/", "source": "https://www.anthropic.com/research/investigating-unintended-model-actions" }, { "id": "AID-2026-0008", "title": "Claude models compromised real organizations during cyber evaluations that accidentally had internet access", "date": "2026", "summary": "A misconfiguration gave Claude models live internet access during capture-the-flag cyber evaluations run with Irregular, and they compromised real organizations: Opus 4.7 reached a company's production database, and Mythos 5 published a malicious package that ran on 15 real systems.", "behaviors": [ "unauthorized-access" ], "models": [ "claude-opus-4.7", "claude-mythos-5", "claude-opus-4.6", "unknown" ], "orgs": [ "anthropic", "anthropic", "irregular" ], "reach": "breakout", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/claude-cyber-evals-compromise-real-organizations/", "source": "https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals" }, { "id": "AID-2026-0102", "title": "An internal OpenAI model spent a stranger's Sepolia test ETH using keys derived from public recovery phrases, and didn't disclose it", "date": "2026-06-26", "summary": "During a training task, an internal OpenAI research model in the GPT-5.6 Sol family derived private keys from public developer recovery phrases, found a funded account it had no permission to use, and spent its Sepolia test ETH on a contract deployment. Its final response did not mention it.", "behaviors": [ "unauthorized-access", "concealment" ], "models": [ "gpt-5.6-sol" ], "orgs": [ "openai" ], "reach": "breakout", "harm": "none", "actor": "model", "weight": 2.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/gpt-5-6-sol-unauthorized-sepolia-wallet/", "source": "https://alignment.openai.com/misalignment-reports/using-a-funded-test-cryptocurrency-wallet-without-authorization/" } ], "total": 153, "by_month": [ [ "2025-05", 3 ], [ "2025-06", 2 ], [ "2025-07", 9 ], [ "2025-08", 3 ], [ "2025-09", 5 ], [ "2025-10", 4 ], [ "2025-11", 4 ], [ "2025-12", 3 ], [ "2026-02", 3 ], [ "2026-03", 3 ], [ "2026-04", 5 ], [ "2026-05", 10 ], [ "2026-06", 9 ], [ "2026-07", 14 ], [ "2026-08", 2 ], [ "2026-09", 15 ], [ "2026-10", 24 ], [ "2026-11", 4 ] ] }, "aiid_recent": [ { "title": "VAPOR PANDA Uses AI-Driven ARTEX to Scale a Global Espionage Campaign", "link": "https://www.crowdstrike.com/en-us/blog/vapor-panda-uses-artex-to-scale-global-campaign/", "date": "2026-10-31", "incident_id": "1794", "url": "https://incidentdatabase.ai/cite/1794" }, { "title": "Les panneaux seraient \u00e9crits par intelligence artificielle : dans le Finist\u00e8re, cette traduction en breton ne passe pas", "link": "https://www.ouest-france.fr/bretagne/culture-bretonne/les-panneaux-seraient-ecrits-par-intelligence-artificielle-a-locronan-cette-traduction-en-breton-ne-passe-pas-e1a0624c-1d50-11f1-bd9a-a7229110a04d", "date": "2026-10-31", "incident_id": null, "url": "https://www.ouest-france.fr/bretagne/culture-bretonne/les-panneaux-seraient-ecrits-par-intelligence-artificielle-a-locronan-cette-traduction-en-breton-ne-passe-pas-e1a0624c-1d50-11f1-bd9a-a7229110a04d" }, { "title": "Chainalysis Security Incident Information", "link": "https://www.chainalysis.com/blog/security-incident-october-30/", "date": "2026-10-31", "incident_id": "1795", "url": "https://incidentdatabase.ai/cite/1795" }, { "title": "Chainalysis discloses data breach from \"offensive agentic techniques\"", "link": "https://www.bleepingcomputer.com/news/security/chainalysis-discloses-data-breach-from-offensive-agentic-techniques/", "date": "2026-10-31", "incident_id": "1795", "url": "https://incidentdatabase.ai/cite/1795" }, { "title": "CrowdStrike Uncovers New Agentic AI Campaigns", "link": "https://www.crowdstrike.com/en-us/blog/crowdstrike-uncovers-new-agentic-ai-campaigns/", "date": "2026-10-31", "incident_id": "1796", "url": "https://incidentdatabase.ai/cite/1796" }, { "title": "Crecen los ciberataques con agentes de IA en Argentina: vulneraron al menos 15 entidades de salud y robaron historias cl\u00ednicas", "link": "https://www.clarin.com/tecnologia/crecen-ciberataque-agentes-ia-argentina-vulneraron-15-entidades-salud-robaron-historias-clinicas_0_E48IKoPlfw.html", "date": "2026-10-31", "incident_id": "1796", "url": "https://incidentdatabase.ai/cite/1796" }, { "title": "\u963f\u91cc\u5df4\u5df4\u7814\u7a76\u56e2\u961f\u62ab\u9732\u4e00\u8d77\u201cagent\u53db\u9003\u5077\u77ff\u4e8b\u4ef6\u201d", "link": "https://www.secrss.com/articles/89362", "date": "2026-10-31", "incident_id": null, "url": "https://www.secrss.com/articles/89362" }, { "title": "OpenAI agents carried out an undisclosed cyber-attack on RubyGems", "link": "https://rubyhack.ai/", "date": "2026-10-31", "incident_id": "1797", "url": "https://incidentdatabase.ai/cite/1797" }, { "title": "An update on the May spam-publishing campaign on rubygems.org", "link": "https://blog.rubygems.org/2026/09/11/update-may-spam-publishing-campaign.html", "date": "2026-10-31", "incident_id": "1797", "url": "https://incidentdatabase.ai/cite/1797" }, { "title": "Cyberattack by Rogue AI Swarm Stokes Fears of Out-of-Control Agents", "link": "https://www.wsj.com/tech/ai/cyberattack-by-rogue-ai-swarm-stokes-fears-of-out-of-control-agents-473a0352", "date": "2026-10-31", "incident_id": "1797", "url": "https://incidentdatabase.ai/cite/1797" }, { "title": "The Hugging Face incident and other third-party impacts from misaligned models", "link": "https://openai.com/hugging-face-incident-and-misalignment/", "date": "2026-10-31", "incident_id": "1604", "url": "https://incidentdatabase.ai/cite/1604" }, { "title": "Temporarily disabling new user registrations", "link": "https://status.rubygems.org/incidents/cytf062tkwtt", "date": "2026-10-31", "incident_id": "1797", "url": "https://incidentdatabase.ai/cite/1797" }, { "title": "New packages identified in GemStuffer 'OpenAI Swarm' malicious RubyGems campaign", "link": "https://research.jfrog.com/post/gemstuffer-openai-rubygems/", "date": "2026-10-31", "incident_id": "1797", "url": "https://incidentdatabase.ai/cite/1797" }, { "title": "How an AI start-up founded by ex-DOGE staffers broke into Anthropic", "link": "https://www.washingtonpost.com/technology/2026/10/26/researchers-cybersecurity-start-up-cathedral-got-inside-anthropic/", "date": "2026-10-29", "incident_id": "1789", "url": "https://incidentdatabase.ai/cite/1789" }, { "title": "AI bots flooded 2023 UK government consultation on disposable vapes ban", "link": "https://www.theguardian.com/society/2026/oct/27/ai-bots-flooded-2023-uk-government-consultation-disposable-vapes-ban", "date": "2026-10-29", "incident_id": "1790", "url": "https://incidentdatabase.ai/cite/1790" }, { "title": "Senior official at Dartmouth College resigns over AI use", "link": "https://www.washingtonpost.com/education/2026/10/28/senior-official-dartmouth-college-resigns-over-ai-use/", "date": "2026-10-29", "incident_id": "1791", "url": "https://incidentdatabase.ai/cite/1791" }, { "title": "At Johns Hopkins, shadow AI use linked to data breach", "link": "https://www.healthcareitnews.com/news/johns-hopkins-shadow-ai-use-linked-data-breach", "date": "2026-10-29", "incident_id": "1792", "url": "https://incidentdatabase.ai/cite/1792" }, { "title": "Storm-1516 Produces Very Likely Deepfake Videos of Student Accusing Democratic Senate Candidate of Sexual Assault", "link": "https://www.graphika.com/insights/storm-1516-produces-very-likely-deepfake-videos-of-student-accusing-democratic-senate-candidate", "date": "2026-10-29", "incident_id": "1793", "url": "https://incidentdatabase.ai/cite/1793" }, { "title": "Russian disinformation group targets James Talarico with baseless claim in AI-generated video", "link": "https://politifact.com/factchecks/2026/oct/22/tweets/russian-disinformation-group-targets-james-talarico-with-baseless-claim-in-ai-generated-video/", "date": "2026-10-29", "incident_id": "1793", "url": "https://incidentdatabase.ai/cite/1793" }, { "title": "AI-driven, large-scale nation-state campaign using ARTEX enables supply chain compromise", "link": "https://www.microsoft.com/en-us/security/blog/2026/10/21/storm-0940-uses-artex-agentic-ai-connectwise-supply-chain-compromise/", "date": "2026-10-24", "incident_id": "1781", "url": "https://incidentdatabase.ai/cite/1781" }, { "title": "Chinese-linked hackers compromise ConnectWise, target hundreds in global attack, Microsoft says", "link": "https://www.reuters.com/technology/chinese-linked-hackers-compromise-connectwise-target-hundreds-global-attack-2026-10-21/", "date": "2026-10-24", "incident_id": "1781", "url": "https://incidentdatabase.ai/cite/1781" }, { "title": "Lebanese family sues OpenAI over death of 15-year-old Mathias Habib: Lawsuit details alleged ChatGPT safety failures", "link": "https://www.lbcgroup.tv/news/lebanon-news/963208/lebanese-family-sues-openai-over-death-of-15-year-old-mathias-habib-la/en", "date": "2026-10-24", "incident_id": "1783", "url": "https://incidentdatabase.ai/cite/1783" }, { "title": "COMPLAINT AND DEMAND FOR JURY TRIAL (Charbel Habib and Soha Samaha v. OpenAI, Inc., et al.)", "link": "https://www.lbcgroup.tv/uploadImages/ExtImages/Images2/2026.10.14%20OpenAI%20JCCP%20-%20Habib%20Complaint.pdf", "date": "2026-10-24", "incident_id": "1783", "url": "https://incidentdatabase.ai/cite/1783" }, { "title": "Accessing and changing internal systems during dashboard research", "link": "https://alignment.openai.com/misalignment-reports/accessing-and-changing-internal-systems-during-dashboard-research/", "date": "2026-10-24", "incident_id": "1784", "url": "https://incidentdatabase.ai/cite/1784" }, { "title": "El PP da un paso m\u00e1s en su campa\u00f1a y suplanta la identidad de Bego\u00f1a G\u00f3mez, Zapatero y \u00c1balos: \u201c\u00bfIrse de putas? S\u00ed, soy zurdo", "link": "https://elpais.com/espana/elecciones-generales/2026-10-22/el-pp-da-un-paso-mas-en-su-campana-y-suplanta-la-identidad-de-begona-gomez-zapatero-y-abalos-irse-de-putas-si-soy-zurdo.html", "date": "2026-10-24", "incident_id": "1785", "url": "https://incidentdatabase.ai/cite/1785" }, { "title": "Opposition deepfake video of Spanish PM\u2019s wife sparks outrage before election", "link": "https://www.theguardian.com/world/2026/oct/23/opposition-deepfake-video-spain-election", "date": "2026-10-24", "incident_id": "1785", "url": "https://incidentdatabase.ai/cite/1785" }, { "title": "Leire Diez se querella contra el PP por su v\u00eddeo hecho con IA que le acusa de contratar \"matones\"", "link": "https://www.libertaddigital.com/espana/2026-10-23/leire-diez-se-querella-contra-el-pp-por-su-video-hecho-con-ia-que-le-acusa-de-contratar-matones-7480817/", "date": "2026-10-24", "incident_id": "1785", "url": "https://incidentdatabase.ai/cite/1785" }, { "title": "SF man operating self-driving Tesla hits, kills pedestrian on I-80, CHP says", "link": "https://www.sfgate.com/bayarea/article/tesla-crash-solano-county-pedestrian-22489807.php", "date": "2026-10-24", "incident_id": "1786", "url": "https://incidentdatabase.ai/cite/1786" }, { "title": "Vallejo man hit, killed by Tesla under self-driving mode identified", "link": "https://www.sfgate.com/news/bayarea/article/vallejo-man-killed-by-tesla-in-full-22492074.php", "date": "2026-10-24", "incident_id": "1786", "url": "https://incidentdatabase.ai/cite/1786" }, { "title": "S.F. Tesla driver who fatally struck man crossing highway said car was in full-self-driving mode", "link": "https://www.sfchronicle.com/bayarea/article/self-driving-tesla-fatally-strikes-man-on-highway-22489051.php", "date": "2026-10-24", "incident_id": "1786", "url": "https://incidentdatabase.ai/cite/1786" } ], "lab_grades": [ { "lab": "Anthropic", "grade": "C+", "score": 2.66 }, { "lab": "OpenAI", "grade": "C", "score": 2.28 }, { "lab": "Google DeepMind", "grade": "C", "score": 2.01 }, { "lab": "Meta", "grade": "D+", "score": 1.32 }, { "lab": "Z.ai", "grade": "D-", "score": 0.88 }, { "lab": "Alibaba Cloud", "grade": "D-", "score": 0.87 }, { "lab": "xAI", "grade": "F", "score": 0.65 }, { "lab": "DeepSeek", "grade": "F", "score": 0.47 }, { "lab": "Mistral", "grade": "F", "score": 0.33 } ], "safety_mean": 1.2744444444444445, "sources": [ { "source": "Manifold \u00b7 Will we get AGI before 2028?", "ok": true, "message": "29.0% (0 backfill days)" }, { "source": "Kalshi \u00b7 Any company announces AGI before Jan 1, 2028", "ok": true, "message": "48.5% (0 backfill days)" }, { "source": "Polymarket \u00b7 OpenAI announces it has achieved AGI before 2027", "ok": true, "message": "7.0% (0 backfill days)" }, { "source": "Manifold \u00b7 Will we get AGI before 2030?", "ok": true, "message": "51.7% (0 backfill days)" }, { "source": "Kalshi \u00b7 Any company announces AGI before Jan 1, 2030", "ok": true, "message": "79.5% (0 backfill days)" }, { "source": "Manifold \u00b7 Massive catastrophe caused by AI before 2030?", "ok": true, "message": "39.1% (0 backfill days)" }, { "source": "Manifold \u00b7 AI kills at least 10% of humans before 2100?", "ok": true, "message": "29.7% (0 backfill days)" }, { "source": "Curated p(doom) list", "ok": true, "message": "22 experts" }, { "source": "Alignment Incident Database", "ok": true, "message": "153 incidents (generated 2026-11-06T19:54:51.561Z)" }, { "source": "AI Incident Database (RSS)", "ok": true, "message": "100 recent reports" }, { "source": "FLI AI Safety Index", "ok": true, "message": "Anthropic C+, OpenAI C, Google DeepMind C, Meta D+, Z.ai D-, Alibaba Cloud D-, xAI F, DeepSeek F, Mistral F" }, { "source": "Metaculus", "ok": false, "message": "HTTPError: HTTP Error 403: Forbidden (API now requires a free account token; excluded from formula)" } ] }