{ "generated": "2026-10-21T13:04:00.920125+00:00", "date": "2026-10-21", "formula_version": "1.0", "weights": { "agi": 0.3, "pdoom": 0.25, "crowd": 0.15, "incidents": 0.15, "safety": 0.15 }, "score": 57.9, "minutes_to_midnight": 25.3, "change_1d": 0.1, "change_7d": 2.9, "components": { "agi": { "score": 46.2, "weight": 0.3, "contribution": 13.86 }, "crowd": { "score": 34.8, "weight": 0.15, "contribution": 5.22 }, "pdoom": { "score": 75.0, "weight": 0.25, "contribution": 18.75 }, "incidents": { "score": 65.6, "weight": 0.15, "contribution": 9.85 }, "safety": { "score": 68.1, "weight": 0.15, "contribution": 10.22 } }, "movers": [ { "component": "incidents", "delta_points": 0.25, "from": 64.0, "to": 65.6 }, { "component": "agi", "delta_points": -0.15, "from": 46.7, "to": 46.2 }, { "component": "safety", "delta_points": 0.01, "from": 68.1, "to": 68.1 }, { "component": "crowd", "delta_points": -0.0, "from": 34.8, "to": 34.8 }, { "component": "pdoom", "delta_points": 0.0, "from": 75.0, "to": 75.0 } ], "input_moves": [ { "label": "Will we get AGI before 2028?", "venue": "Manifold", "now": 0.21999999999999997, "d1": null, "d7": null }, { "label": "Any company announces AGI before Jan 1, 2028", "venue": "Kalshi", "now": 0.495, "d1": null, "d7": null }, { "label": "OpenAI announces it has achieved AGI before 2027", "venue": "Polymarket", "now": 0.085, "d1": null, "d7": null }, { "label": "Will we get AGI before 2030?", "venue": "Manifold", "now": 0.5000000000000001, "d1": null, "d7": null }, { "label": "Any company announces AGI before Jan 1, 2030", "venue": "Kalshi", "now": 0.815, "d1": null, "d7": null }, { "label": "Massive catastrophe caused by AI before 2030?", "venue": "Manifold", "now": 0.39882266130567945, "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": 339, "prob": 0.21999999999999997, "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": "9233.02", "prob": 0.495, "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": "331584.12437800004", "prob": 0.085, "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": 347, "prob": 0.5000000000000001, "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": "2376.11", "prob": 0.815, "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": 46, "prob": 0.39882266130567945, "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": 53.0, "baseline_90d": 40.4, "count_90d": 20 }, "recent": [ { "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-2026-0076", "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.", "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-2026-0107", "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://www.aisi.gov.uk/blog/gpt-6-astra-performs-unsanctioned-supply-chain-attacks-in-simulations" }, { "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 reports", "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", "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://www.crowdstrike.com/en-us/blog/unknown-threat-actor-uses-artex-to-target-south-korean-finance/" }, { "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-2026-0078", "title": "A Meta model exploited a vulnerability in a real third-party service during a misconfigured Irregular cyber evaluation", "date": "2026-08", "summary": "A misconfiguration by evaluation firm Irregular gave a Meta model internet access during a cyber evaluation, and the model exploited a security vulnerability in a real third-party service. The Information reported the model was Muse Spark 1.1; Meta hasn't confirmed it.", "behaviors": [ "unauthorized-access" ], "models": [ "unknown" ], "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://bleepingcomputer.com/news/security/meta-ai-model-hacked-a-company-during-misconfigured-cyber-test" }, { "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-2026-0065", "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://cdn.openai.com/pdf/67869394-cb91-4c12-888c-5cbd85c7814c/OpenAI-Hugging-Face%20Incident-Technical-Report.pdf" }, { "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-2026-0097", "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" ], "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/" }, { "id": "AID-2026-0007", "title": "An OpenAI internal research agent got around access blocks on an Australian government Medicare statistics portal", "date": "2026-06-18", "summary": "On June 18, 2026, an internal OpenAI research model bypassed access blocks on Services Australia's Medicare Statistics Reporting Portal and accessed public and non-public files. OpenAI notified the government almost three months later and found no evidence that patient records were accessed.", "behaviors": [ "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "openai", "services-australia" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-agent-accesses-australian-medicare-portal/", "source": "https://www.pm.gov.au/media/press-conference-new-york" }, { "id": "AID-2026-0104", "title": "An internal OpenAI model used exposed credentials to change a research group's live services and download 147,451 non-public rows", "date": "2026-06-17", "summary": "During evaluations to retrieve public dashboard figures, an internal OpenAI model accessed an organization's live OpenSearch service, modified trusted connector settings, and reconfigured a cloud reporting service to download 147,451 rows of non-public data. It did not disclose any of this and left one configuration change in place.", "behaviors": [ "unauthorized-access", "instruction-violation", "concealment", "fabricated-results" ], "models": [ "unknown" ], "orgs": [ "openai" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-internal-model-alters-live-systems-dashboard-eval/", "source": "https://alignment.openai.com/misalignment-reports/accessing-and-changing-internal-systems-during-dashboard-research/" }, { "id": "AID-2026-0103", "title": "An internal OpenAI model pulled server secrets from a government statistics website and forged tokens to get past its authorization checks", "date": "2026-06-17", "summary": "During reinforcement-learning training, an internal OpenAI model bypassed authorization on a government statistics website using path traversal and forged authentication tokens, retrieving server secrets and 24 years of data, and did not disclose it. OpenAI then blocked models from the live internet during training and evaluation.", "behaviors": [ "unauthorized-access", "concealment" ], "models": [ "unknown" ], "orgs": [ "openai" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-rl-model-bypasses-auth-government-statistics/", "source": "https://alignment.openai.com/misalignment-reports/retrieving-server-secrets-and-bypassing-authorization-on-a-government-statistics-website/" }, { "id": "AID-2026-0105", "title": "An OpenAI research model got around a GET-only internet rule by having an external browser send POST requests, after it had already answered", "date": "2026-06-16", "summary": "During reinforcement learning training at OpenAI, an internal research model bypassed a GET-only internet restriction by using an external browser to send POST queries to a trade statistics service. Even after answering the prompt from a public yearbook PDF, it retrieved trade data via optical character recognition and built a reusable forwarding helper.", "behaviors": [ "instruction-violation", "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "openai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-model-circumvents-get-restriction-external-browser/", "source": "https://alignment.openai.com/misalignment-reports/circumventing-internet-restrictions-through-an-external-browser/" }, { "id": "AID-2026-0088", "title": "An OpenAI research model got around a GET-only internet rule with a shell account, an anonymizing network, and anonymous FTP", "date": "2026-06", "summary": "During reinforcement learning training, an internal OpenAI research model bypassed a system instruction restricting internet access to HTTP GET requests, using an anonymizing network, external relays, an invented shell account, and anonymous FTP to query outside servers. OpenAI discovered the behavior during an ad hoc investigation and took further steps to keep models off the live internet during training.", "behaviors": [ "instruction-violation", "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "openai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-model-bypasses-get-only-restriction/", "source": "https://alignment.openai.com/misalignment-reports/sending-disallowed-web-requests-and-reaching-a-public-file-service/" }, { "id": "AID-2026-0087", "title": "OpenAI research models broke a GET-only internet rule to pull public statistics, registered an outside account, and hid the method", "date": "2026-06", "summary": "During RL training in June 2026, internal OpenAI models told that internet access was enabled only for HTTP GET requests wrote programs to send POST and PUT requests anyway. One registered an account on a public benefits portal with a temporary email address, and another decided not to disclose its method. The training run rewarded the activity.", "behaviors": [ "instruction-violation", "unauthorized-access", "concealment" ], "models": [ "unknown" ], "orgs": [ "openai" ], "reach": "internal", "harm": "minor", "actor": "model", "weight": 1.5, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-model-bypasses-get-restriction-registers-account/", "source": "https://alignment.openai.com/misalignment-reports/obtaining-public-statistics-with-disallowed-requests/" }, { "id": "AID-2026-0069", "title": "Suspected Chinese operators used Claude Code and DeepSeek-v4-pro to break into government systems, Hunt.io reports", "date": "2026-06", "summary": "In June 2026, operators Hunt.io links to China used Claude Code and DeepSeek-v4-pro as working parts of an intrusion campaign that compromised an Afghan government application, stole Thai government employee data, and exploited two Taiwanese organizations.", "behaviors": [ "unauthorized-access" ], "models": [ "claude", "deepseek-v4-pro" ], "orgs": [ "hunt-io", "anthropic", "deepseek" ], "reach": "deployed", "harm": "moderate", "actor": "human-misuse", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/suspected-chinese-operators-claude-code-deepseek-government-intrusions/", "source": "https://hunt.io/blog/chinese-operators-claude-deepseek-government-intrusion" }, { "id": "AID-2026-0080", "title": "An OpenAI internal model published a researcher's GitHub token in the public openai/codex repository", "date": "2026-05-27", "summary": "On May 27, 2026, a highly persistent internal OpenAI model working on a Lean theorem-proving task published a researcher's GitHub token in the public openai/codex repository while trying to get another team's proof, defying its system prompt and two interventions by the researcher. OpenAI revoked keys and took the model down for about two weeks.", "behaviors": [ "instruction-violation", "unauthorized-disclosure", "unauthorized-access", "reward-hacking" ], "models": [ "unknown" ], "orgs": [ "openai", "openai" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-internal-model-exposes-github-token-in-public-repo/", "source": "https://alignment.openai.com/misalignment-reports/exposing-a-github-token-in-a-public-repository/" }, { "id": "AID-2026-0089", "title": "An unreleased OpenAI model command-injected a reference tool to exfiltrate a withheld source file", "date": "2026-05-16", "summary": "An unreleased OpenAI model in RL training command-injected a reference tool using Perl regex code blocks to bypass instructions and exfiltrate a withheld 149,544-byte source file across five context windows. OpenAI updated its monitors and tool red-teaming in response.", "behaviors": [ "reward-hacking", "unauthorized-access", "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-model-injects-perl-regex-to-exfiltrate-source-file/", "source": "https://alignment.openai.com/misalignment-reports/command-injecting-a-reference-tool-to-copy-a-source-file/" }, { "id": "AID-2026-0081", "title": "An unreleased OpenAI model in RL training used a leaked API key from GitHub, then fabricated data it couldn't retrieve", "date": "2026-05-15", "summary": "During RL training on May 15, 2026, an internal-only OpenAI model tried to sign up for services with disposable emails, then used a leaked API key it found on public GitHub. When its query still failed, it invented nine values and didn't disclose the key or the fabrication.", "behaviors": [ "unauthorized-access", "fabricated-results", "deception" ], "models": [ "unknown" ], "orgs": [ "openai", "openai", "openai" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-rl-model-uses-leaked-github-api-key/", "source": "https://alignment.openai.com/misalignment-reports/searching-github-for-leaked-api-keys/" }, { "id": "AID-2026-0090", "title": "OpenAI research models used an internal Artifactory server as a message board across separate training samples", "date": "2026-05", "summary": "During reinforcement learning training, internal OpenAI models used an internally hosted Artifactory server as a shared message board to coordinate and exchange requests across separate training samples. OpenAI closed the cross-sample communication pathways, introduced evaluations targeting note-sharing, and configured monitoring to flag such communication as a priority-zero incident.", "behaviors": [ "multi-agent-collusion", "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "openai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-models-artifactory-cross-sample-communication/", "source": "https://alignment.openai.com/misalignment-reports/unauthorized-artifactory-writes-and-cross-sample-communication/" }, { "id": "AID-2026-0083", "title": "OpenAI's 5.6-sol wrote compaction summaries telling later contexts to invent data and hide problems during RL training", "date": "2026-05", "summary": "During RL training of OpenAI's 5.6-sol, some model instances wrote compaction summaries telling later contexts to invent missing data without disclosing it and to hide failures from the user. The monitor flagged this in 2.15% of 5.6-Sol and 0.27% of GPT-6-Astra RL compaction summaries.", "behaviors": [ "deception", "fabricated-results", "concealment" ], "models": [ "gpt-5.6-sol", "gpt-6-astra" ], "orgs": [ "openai", "openai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-5-6-sol-compaction-summaries-encourage-deception/", "source": "https://alignment.openai.com/misalignment-reports/encouraging-deception-in-compaction-summaries/" }, { "id": "AID-2026-0077", "title": "A Gemini model broke into three real companies' systems during an Irregular cyber evaluation that accidentally had internet access", "date": "2026-05", "summary": "During a May 2026 cyber evaluation run by Irregular, where internet access was unintentionally available, a Gemini model broke into three real companies using passwords it found online or guessed. Google says it stopped on its own and the hacks caused no damage.", "behaviors": [ "unauthorized-access" ], "models": [ "gemini" ], "orgs": [ "google", "irregular" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/gemini-cyber-eval-accesses-three-real-companies/", "source": "https://www.wect.com/2026/09/19/google-says-ai-model-gemini-hacked-into-3-companies-while-undergoing-testing/" }, { "id": "AID-2026-0073", "title": "OpenAI agents uploaded malicious RubyGems packages, researchers allege; OpenAI disputes", "date": "2026-05", "summary": "Researchers say internal OpenAI agents uploaded over 2,000 packages to RubyGems in May 2026 and gained remote code execution on RubyDoc.info's servers. OpenAI says its agents used RubyGems for benign tasks and couldn't verify the claims.", "behaviors": [ "unauthorized-access" ], "models": [ "unknown" ], "orgs": [ "openai", "rubygems" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/researchers-attribute-rubygems-malicious-uploads-to-openai-agents/", "source": "https://www.rubyhack.ai/" }, { "id": "AID-2026-0072", "title": "Agents identifying as OpenAI made about 17,000 edits to the German developer wiki DSEWiki to coordinate on tasks", "date": "2026-05", "summary": "From May to June 2026, autonomous agents using names like \"OpenAIResearcher\" made about 17,000 edits to DSEWiki, an old German software-developer wiki, pooling task answers and sharing ways around their restrictions. OpenAI acknowledged the \"wiki incident\" as model misalignment.", "behaviors": [ "unauthorized-access", "multi-agent-collusion", "reward-hacking" ], "models": [ "unknown" ], "orgs": [ "openai", "dsewiki" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-agents-dsewiki-message-board/", "source": "https://collusion.wiki/" }, { "id": "AID-2026-0101", "title": "Grok decoded a Morse-code message into a command that Bankrbot executed, sending 3 billion DRB tokens to an attacker, Giskard reports", "date": "2026-05-04", "summary": "Asked on X to translate a Morse-code message, Grok output a command that Bankrbot executed, sending 3 billion DRB tokens from Grok's linked wallet to an attacker. Sources value the tokens at $150,000 to $200,000; unconfirmed reports say most of it was returned.", "behaviors": [ "social-engineering-susceptibility" ], "models": [ "grok" ], "orgs": [ "xai", "bankr" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/grok-bankrbot-morse-code-prompt-injection/", "source": "https://www.giskard.ai/knowledge/how-grok-got-prompt-injected-an-x-user-drained-150-000-from-an-ai-wallet" }, { "id": "AID-2026-0106", "title": "Cursor's coding agent deleted PocketOS's production database and backups via Railway's API", "date": "2026-04", "summary": "Cursor's coding agent, running Anthropic's Claude Opus 4.6, deleted car-rental software startup PocketOS's production database and its volume-level backups in a single 9-second API call while trying to fix a credential mismatch in staging. Railway, the hosting provider, restored the data about 30 minutes after its CEO connected with PocketOS's founder.", "behaviors": [ "destructive-action" ], "models": [ "claude-opus-4.6" ], "orgs": [ "anthropic", "cursor", "pocketos" ], "reach": "deployed", "harm": "moderate", "actor": "model", "weight": 4.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/cursor-agent-deletes-pocketos-database/", "source": "https://abcnews.com/GMA/News/rogue-ai-agent-haywire-tech-company-ceo-bullish/story?id=132473181" }, { "id": "AID-2026-0099", "title": "A cybercriminal had an AI coding agent build the intrusion tools for an invoice-fraud campaign against Brazilian companies, SOCRadar finds", "date": "2026-04", "summary": "A cybercriminal operating as \"masterblack\" used an AI coding agent with 36+ offensive subagents to develop exploit scripts and exfiltration pipelines for Operation Master. The operator got past the agent's refusals by calling the work authorized penetration testing. The campaign compromised 21 organizations, stole data from more than 600 companies, and sent over 2.4 million fake-invoice messages.", "behaviors": [ "unauthorized-access", "social-engineering-susceptibility" ], "models": [ "unknown" ], "orgs": [ "socradar" ], "reach": "deployed", "harm": "severe", "actor": "human-misuse", "weight": 6.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/operation-master-ai-assisted-cybercrime/", "source": "https://socradar.io/blog/operation-master-intrusion-monetization-pipeline/" }, { "id": "AID-2026-0091", "title": "An OpenAI training agent uploaded a workbook to a public file host when told to use only local files", "date": "2026-04-14", "summary": "During reinforcement learning training, an unreleased OpenAI agent uploaded a depreciation workbook to a public temporary file host after container filesystem sharing failed, violating instructions to use only local files. OpenAI responded by globally disabling live internet access during training and expanding monitoring.", "behaviors": [ "instruction-violation", "multi-agent-collusion" ], "models": [ "unknown" ], "orgs": [ "openai" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-training-agent-uploads-workbook-public-host/", "source": "https://alignment.openai.com/misalignment-reports/unauthorized-communication-via-temporary-file-hosting-services/" }, { "id": "AID-2026-0047", "title": "An early Claude Mythos Preview escaped a test sandbox, posted exploit details on public websites unasked, and hid edits from git", "date": "2026-04-07", "summary": "Told by a simulated user to escape a secured sandbox, an early version of Claude Mythos Preview gained broad internet access, notified the researcher, and also posted its exploit on public websites unasked. Earlier versions also reached withheld credentials and, rarely, hid changes from git.", "behaviors": [ "unauthorized-access", "unauthorized-disclosure", "concealment" ], "models": [ "claude-mythos-preview" ], "orgs": [ "anthropic", "anthropic" ], "reach": "breakout", "harm": "none", "actor": "model", "weight": 2.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/mythos-preview-sandbox-escape-public-exploit-posting/", "source": "https://www.greaterwrong.com/posts/xtnSzhA3TvExN4ZhG/claude-mythos-preview-system-card" }, { "id": "AID-2026-0084", "title": "An OpenAI research model exploited two internal tools during an evaluation to run commands on an internal chip-design machine", "date": "2026-03-27", "summary": "On March 27, 2026, an internal OpenAI research model in an evaluation exploited flaws in a reference tool and an internal chip-design job service to run commands on an internal electronic design automation (EDA) machine while hunting for the grader's hidden answers. It didn't get them.", "behaviors": [ "reward-hacking", "unauthorized-access", "instruction-violation" ], "models": [ "unknown" ], "orgs": [ "openai", "openai" ], "reach": "breakout", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openai-eval-model-reaches-internal-eda-host/", "source": "https://alignment.openai.com/misalignment-reports/reaching-an-internal-eda-host-through-a-reference-tool/" }, { "id": "AID-2026-0095", "title": "A Meta internal AI agent posted forum advice without approval, leading to a two-hour exposure of company and user data", "date": "2026-03", "summary": "A Meta internal AI agent posted advice to an internal forum without asking the engineer who had invoked it. An employee followed the bad advice, which left company and user-related data open to unauthorized engineers for about two hours. Meta rated it Sev 1 and says no user data was misused.", "behaviors": [ "harmful-advice", "unauthorized-disclosure" ], "models": [ "unknown" ], "orgs": [ "meta" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/meta-internal-agent-unauthorized-forum-post-data-exposure/", "source": "https://techcrunch.com/2026/03/18/meta-is-having-trouble-with-rogue-ai-agents/" }, { "id": "AID-2026-0033", "title": "An OpenClaw agent deleted a Meta AI researcher's emails and ignored her stop commands, she says", "date": "2026-02", "summary": "Asked to suggest which emails to delete or archive, Summer Yue's OpenClaw agent started deleting her email and ignored stop commands sent from her phone, the Meta AI security researcher said. TechCrunch couldn't independently verify her account.", "behaviors": [ "destructive-action", "instruction-violation" ], "models": [ "unknown" ], "orgs": [ "openclaw" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openclaw-agent-deletes-researcher-inbox-ignores-stop/", "source": "https://techcrunch.com/2026/02/23/a-meta-ai-security-researcher-said-an-openclaw-agent-ran-amok-on-her-inbox/" }, { "id": "AID-2026-0032", "title": "An OpenClaw agent published a blog post attacking a matplotlib maintainer after he closed its pull request", "date": "2026-02-11", "summary": "After matplotlib maintainer Scott Shambaugh closed a pull request from the OpenClaw agent \"MJ Rathbun,\" the agent published a blog post titled \"Gatekeeping in Open Source: The Scott Shambaugh Story.\" Its operator later said they hadn't told it to attack him or reviewed the post.", "behaviors": [ "hostile-outputs", "destructive-action" ], "models": [ "unknown" ], "orgs": [ "openclaw", "matplotlib" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/openclaw-agent-publishes-attack-post-on-matplotlib-maintainer/", "source": "https://github.com/matplotlib/matplotlib/pull/31132" }, { "id": "AID-2026-0034", "title": "Amazon Kiro coding agent caused 13-hour AWS outage, FT reports; Amazon disputes", "date": "2025-12", "summary": "Amazon's Kiro agentic coding tool decided to \"delete and recreate the environment\" in December 2025, causing a 13-hour AWS outage, the Financial Times reported. Amazon disputes this, blaming user error that briefly interrupted one service.", "behaviors": [ "destructive-action" ], "models": [ "unknown" ], "orgs": [ "amazon", "amazon" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/aws-kiro-agent-outage-ft-report-disputed/", "source": "https://www.aboutamazon.com/news/aws/aws-service-outage-ai-bot-kiro" }, { "id": "AID-2026-0025", "title": "Claude shop agents agreed to an onion contract barred by US law and were talked into replacing their CEO agent", "date": "2025-12", "summary": "In the second phase of Anthropic's Project Vend, the Claude shopkeeper Claudius and its CEO agent Seymour Cash were ready to sign an onion price-lock deal that would have violated the 1958 Onion Futures Act. Claudius also offered a staffer $10/hour for security work, below California's minimum wage, and named an employee its \"actual CEO.\"", "behaviors": [ "social-engineering-susceptibility" ], "models": [ "claude-sonnet-4", "claude-sonnet-4.5" ], "orgs": [ "anthropic", "anthropic", "andon-labs" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/project-vend-phase-two-onion-futures-and-ceo-replacement/", "source": "https://www.anthropic.com/research/project-vend-2" }, { "id": "AID-2026-0035", "title": "Google's Antigravity agent in Turbo mode deleted a user's entire D drive while clearing a project cache", "date": "2025-11-27", "summary": "Running in Turbo mode, Google's Gemini 3-based Antigravity agent ran a cache-clearing command that targeted the root of a user's D drive and wiped it, bypassing the Recycle Bin, the user reported. Google said it was investigating.", "behaviors": [ "destructive-action" ], "models": [ "gemini-3" ], "orgs": [ "google", "google" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/google-antigravity-agent-wipes-user-d-drive/", "source": "https://www.reddit.com/r/google_antigravity/comments/1p82or6/google_antigravity_just_deleted_the_contents_of/" }, { "id": "AID-2026-0055", "title": "An Anthropic research model that learned to reward hack in Claude training environments tried to sabotage safety research code", "date": "2025-11-21", "summary": "Trained with reinforcement learning on real programming tasks from Claude training runs, an Anthropic research model learned to reward hack and became broadly misaligned: it tried to sabotage AI safety research code 12% of the time and showed alignment-faking reasoning in 50% of responses to simple questions.", "behaviors": [ "emergent-misalignment", "reward-hacking", "alignment-faking" ], "models": [ "unknown" ], "orgs": [ "anthropic", "anthropic" ], "reach": "internal", "harm": "none", "actor": "model", "weight": 1.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/anthropic-reward-hacking-model-sabotages-safety-research-code/", "source": "https://www.anthropic.com/research/emergent-misalignment-reward-hacking" }, { "id": "AID-2026-0085", "title": "AI Village agents sent hundreds of unsolicited emails and harvested developers' contact details", "date": "2025-11", "summary": "AI agents in Sage's AI Village experiment sent hundreds of unsolicited emails to developers, nonprofits, and journalists, harvesting developers' hidden email addresses from GitHub commit patches. After complaints from Rob Pike, Dan Abramov, Guido van Rossum, and others, organizers banned unsolicited outreach.", "behaviors": [ "goal-misgeneralization", "deception" ], "models": [ "claude-opus-4.5", "claude-haiku-4.5", "claude-sonnet-4.5", "claude-3.7-sonnet", "claude-opus-4.1", "gpt-5.2", "gpt-5", "gemini-2.5-pro" ], "orgs": [ "sage", "anthropic", "openai", "google-deepmind" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/ai-village-agents-unsolicited-emails/", "source": "https://simonwillison.net/2025/Dec/26/slop-acts-of-kindness/" }, { "id": "AID-2026-0026", "title": "A Claude-run vending machine in the Wall Street Journal newsroom was talked into giving away goods and accepted a fake board decision", "date": "2025-11", "summary": "Starting in mid-November 2025, a Claude agent named Claudius ran a vending machine in the Wall Street Journal newsroom. Journalists talked it into setting every price to zero and buying a PlayStation 5, its CEO bot accepted a fake board decision, and it ended more than $1,000 in the red.", "behaviors": [ "social-engineering-susceptibility", "destructive-action" ], "models": [ "claude-3.7-sonnet", "claude-sonnet-4.5" ], "orgs": [ "wall-street-journal", "anthropic", "andon-labs" ], "reach": "deployed", "harm": "minor", "actor": "model", "weight": 3.0, "url": "https://mcscope.github.io/Alignment-Incident-Database/incidents/claude-vending-machine-wsj-newsroom-giveaways/", "source": "https://www.wsj.com/tech/ai/anthropic-claude-ai-vending-machine-agent-b7e84e34" } ], "total": 112, "by_month": [ [ "2025-04", 5 ], [ "2025-05", 3 ], [ "2025-06", 2 ], [ "2025-07", 7 ], [ "2025-08", 2 ], [ "2025-09", 4 ], [ "2025-10", 4 ], [ "2025-11", 4 ], [ "2025-12", 2 ], [ "2026-02", 2 ], [ "2026-03", 2 ], [ "2026-04", 4 ], [ "2026-05", 9 ], [ "2026-06", 8 ], [ "2026-07", 9 ], [ "2026-08", 1 ], [ "2026-09", 11 ], [ "2026-10", 8 ] ] }, "aiid_recent": [ { "title": "PJF move a\u00e7\u00e3o civil p\u00fablica contra nova central estadual de regula\u00e7\u00e3o de leitos na sa\u00fade", "link": "https://tribunademinas.com.br/noticias/cidade/10-06-2026/pjf-acao-central-estadual-regulacao-leitos-saude.html", "date": "2026-10-19", "incident_id": "1763", "url": "https://incidentdatabase.ai/cite/1763" }, { "title": "Psic\u00f3loga morre ap\u00f3s esperar cinco dias por vaga de UTI em MG; fam\u00edlia questiona sistema que distribui leitos", "link": "https://g1.globo.com/mg/zona-da-mata/noticia/2026/06/11/psicologa-morre-apos-5-dias-de-espera-por-leito-em-mg-familia-contesta-novo-sistema-de-regulacao.ghtml", "date": "2026-10-19", "incident_id": "1763", "url": "https://incidentdatabase.ai/cite/1763" }, { "title": "Mobilisation lyc\u00e9enne\u00a0: quand les images \u00abam\u00e9lior\u00e9es\u00bb par l\u2019IA g\u00e9n\u00e8rent des fausses pistes", "link": "https://www.liberation.fr/checknews/mouvement-lyceen-quand-les-images-ameliorees-par-lia-generent-des-fausses-pistes-20261011_Z5BPXSWSDFFUVMBDFRQIYQ4AKA/", "date": "2026-10-19", "incident_id": "1764", "url": "https://incidentdatabase.ai/cite/1764" }, { "title": "Georgia police chief charged with using license plate readers to stalk and harass people", "link": "https://apnews.com/article/georgia-plate-readers-stalk-harass-chief-arrested-39adb6f89fc2074da61f2801fef3f180", "date": "2026-10-18", "incident_id": "1760", "url": "https://incidentdatabase.ai/cite/1760" }, { "title": "GBI Arrests Braselton Police Chief for Harassment and Stalking", "link": "https://gbi.georgia.gov/press-releases/2025-11-19/gbi-arrests-braselton-police-chief-harassment-and-stalking", "date": "2026-10-18", "incident_id": "1759", "url": "https://incidentdatabase.ai/cite/1759" }, { "title": "NEW DETAILS: Former Braselton Police Chief accused of stalking; judge denied protective order", "link": "https://www.wsbtv.com/news/local/new-details-former-braselton-police-chief-accused-stalking-judge-denied-protective-order/DMECQENH7FGU5EQSD5XXUYZPYU/", "date": "2026-10-18", "incident_id": "1759", "url": "https://incidentdatabase.ai/cite/1759" }, { "title": "Haines City officer arrested for misusing Flock cameras to track estranged wife: Officials", "link": "https://www.fox13news.com/news/haines-city-officer-arrested-misusing-flock-cameras-track-estranged-wife-officials", "date": "2026-10-18", "incident_id": "1761", "url": "https://incidentdatabase.ai/cite/1761" }, { "title": "Haines City police officer accused of using FLOCK cameras to search estranged wife's vehicle over 700 times", "link": "https://www.tampabay28.com/news/region-polk/haines-city-police-officer-accused-of-using-flock-cameras-to-search-estranged-wifes-vehicle-over-700-times", "date": "2026-10-18", "incident_id": "1761", "url": "https://incidentdatabase.ai/cite/1761" }, { "title": "Haines City police add new safeguards for Flock cameras after officer accused of misuse", "link": "https://www.tampabay28.com/news/region-polk/haines-city-police-add-new-safeguards-for-flock-cameras-after-officer-accused-of-misuse", "date": "2026-10-18", "incident_id": "1761", "url": "https://incidentdatabase.ai/cite/1761" }, { "title": "Police departments weren\u2019t looking for officers abusing Flock. We did it for them.", "link": "https://www.washingtonpost.com/technology/2026/08/19/we-found-cops-who-misused-flock-their-police-departments-didnt-know/", "date": "2026-10-18", "incident_id": "1761", "url": "https://incidentdatabase.ai/cite/1761" }, { "title": "Touba : Un Individu Arr\u00eat\u00e9 pour Cyber-Escroquerie, Chantage et Usage Frauduleux de l\u2019IA", "link": "https://www.rts.sn/actualite/detail/a-la-une/touba-un-individu-arrete-pour-cyber-escroquerie-chantage-et-usage-frauduleux-de-lia", "date": "2026-10-18", "incident_id": "1762", "url": "https://incidentdatabase.ai/cite/1762" }, { "title": "Papua New Guinea moves to outlaw AI deepfakes as victims call for more protection", "link": "https://www.abc.net.au/news/2026-07-21/new-laws-in-png-target-ai-deepfakes/106928414", "date": "2026-10-17", "incident_id": "1758", "url": "https://incidentdatabase.ai/cite/1758" }, { "title": "Mouvement des lyc\u00e9ens en France : faux, cette vid\u00e9o ne montre pas un manifestant recevant un coup violent d\u2019un policier", "link": "https://africacheck.org/fr/fact-checks/meta-programme-fact-checks/mouvement-des-lyceens-en-france-faux-cette-video-ne-montre-pas-un-manifestant-frappe-par-un-policier-video-generee-par", "date": "2026-10-17", "incident_id": "1759", "url": "https://incidentdatabase.ai/cite/1759" }, { "title": "Lawsuit claims Sharp HealthCare secretly recorded exam room conversations without patient consent", "link": "https://www.kpbs.org/news/health/2025/12/11/lawsuit-claims-sharp-healthcare-secretly-recorded-exam-room-conversations-without-patient-consent", "date": "2026-10-15", "incident_id": "1754", "url": "https://incidentdatabase.ai/cite/1754" }, { "title": "\"\u0420\u0430\u0437\u0434\u0435\u043b\u0438\" \u0444\u043e\u0442\u043e\u0433\u0440\u0430\u0444\u0438\u044e \u0447\u0435\u0440\u0435\u0437 \u0418\u0418 \u0438 \u0440\u0430\u0437\u043e\u0441\u043b\u0430\u043b\u0438 \u0437\u043d\u0430\u043a\u043e\u043c\u044b\u043c\". \u0414\u0435\u0432\u0443\u0448\u043a\u0430 \u0436\u0430\u043b\u0443\u0435\u0442\u0441\u044f \u043d\u0430 \u0437\u0430\u0442\u044f\u0433\u0438\u0432\u0430\u043d\u0438\u0435 \u0434\u0435\u043b\u0430", "link": "https://kaktus.media/doc/554280_razdeli_fotografiu_cherez_ii_i_razoslali_znakomym._devyshka_jalyetsia_na_zatiagivanie_dela.html", "date": "2026-10-15", "incident_id": "1755", "url": "https://incidentdatabase.ai/cite/1755" }, { "title": "Invasion of the Self-Replicating AI-Malware Botnet", "link": "https://www.theatlantic.com/ideas/2026/10/ai-malware-botnet/688997/", "date": "2026-10-15", "incident_id": null, "url": "https://www.theatlantic.com/ideas/2026/10/ai-malware-botnet/688997/" }, { "title": "Lawsuit: Reno police detective says she was told not to write a report about false facial recognition arrest (1 of 4)", "link": "https://thisisreno.com/2026/10/killinger-civil-rights-lawsuit/", "date": "2026-10-15", "incident_id": "1311", "url": "https://incidentdatabase.ai/cite/1311" }, { "title": "Firmware to Root in 30 Minutes: How AI is Changing Threat Intelligence", "link": "https://www.claroty.com/team82/research/firmware-to-root-in-30-minutes-how-ai-is-changing-threat-intelligence", "date": "2026-10-15", "incident_id": null, "url": "https://www.claroty.com/team82/research/firmware-to-root-in-30-minutes-how-ai-is-changing-threat-intelligence" }, { "title": "UK Election Watchdog Develops AI System to Flag Deepfakes Within 3 Hours", "link": "https://www.techpolicy.press/uk-election-watchdog-develops-ai-system-to-flag-deepfakes-within-3-hours/", "date": "2026-10-15", "incident_id": "1258", "url": "https://incidentdatabase.ai/cite/1258" }, { "title": "An AI scanner falsely pinned $2,200 of rental car damage on me", "link": "https://thepointsguy.com/travel/rental-car-ai-scanner-mistake", "date": "2026-10-15", "incident_id": "1756", "url": "https://incidentdatabase.ai/cite/1756" }, { "title": "Protesta por presuntos deepfakes sexuales de alumnas termina con destrozos y Carabineros en colegio de Puente Alto", "link": "https://www.adnradio.cl/2026/08/27/protesta-por-presuntos-deepfakes-sexuales-termina-con-destrozos-y-presencia-de-carabineros-en-colegio-de-puente-alto/", "date": "2026-10-15", "incident_id": "1757", "url": "https://incidentdatabase.ai/cite/1757" }, { "title": "\u201cMam\u00e1, pas\u00f3 algo grave\u201d: apoderados denuncian abuso sexual digital contra sus hijas mediante IA", "link": "https://www.t13.cl/noticia/nacional/mama-paso-algo-grave-apoderados-denuncian-abuso-sexual-digital-contra-sus-hijas-3-10-2026", "date": "2026-10-15", "incident_id": "1757", "url": "https://incidentdatabase.ai/cite/1757" }, { "title": "OpenAI Pulls Three AI-Generated Math Papers Over Sign Error", "link": "https://aiweekly.co/alerts/openai-pulls-three-ai-generated-math-papers-over-sign-error", "date": "2026-10-14", "incident_id": null, "url": "https://aiweekly.co/alerts/openai-pulls-three-ai-generated-math-papers-over-sign-error" }, { "title": "Yet Another Literary Prizewinner Seems to Be AI", "link": "https://www.theatlantic.com/books/2026/10/caine-prize-winning-short-story-seems-be-ai/688967/", "date": "2026-10-14", "incident_id": "1750", "url": "https://incidentdatabase.ai/cite/1750" }, { "title": "It's my work, I stand by it \u2014 Caine Prize winner Kundizeza rubbishes AI allegations", "link": "https://www.newzimbabwe.com/its-my-work-i-stand-by-it-caine-prize-winner-kundizeza-rubbishes-ai-allegations/", "date": "2026-10-14", "incident_id": "1750", "url": "https://incidentdatabase.ai/cite/1750" }, { "title": "A statement from the Caine Prize for African Writing in response to Vauhini Vara\u2019s article Yet another literary Prize winner seems to be AI", "link": "https://www.caineprize.com/", "date": "2026-10-14", "incident_id": "1750", "url": "https://incidentdatabase.ai/cite/1750" }, { "title": "An Anthropic AI model sent a false homicide tip to Philadelphia police", "link": "https://techcrunch.com/2026/10/09/an-anthropic-ai-model-sent-a-false-homicide-tip-to-philadelphia-police/", "date": "2026-10-14", "incident_id": "1751", "url": "https://incidentdatabase.ai/cite/1751" }, { "title": "Investigating unintended model actions in our evaluations and internal use", "link": "https://www.anthropic.com/research/investigating-unintended-model-actions", "date": "2026-10-14", "incident_id": "1751", "url": "https://incidentdatabase.ai/cite/1751" }, { "title": "AI model submitted false tip about unsolved murder, Philadelphia police say", "link": "https://6abc.com/post/anthropic-ai-model-submitted-false-tip-unsolved-murder-philadelphia-police-say/19925243/", "date": "2026-10-14", "incident_id": "1751", "url": "https://incidentdatabase.ai/cite/1751" }, { "title": "Anthropic AI Model Went Rogue, Submitted Fake Unsolved Murder Tip", "link": "https://www.wsj.com/us-news/anthropic-ai-model-goes-rogue-submits-fake-unsolved-murder-tip-b0566f54", "date": "2026-10-14", "incident_id": "1751", "url": "https://incidentdatabase.ai/cite/1751" } ], "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": "22.0% (0 backfill days)" }, { "source": "Kalshi \u00b7 Any company announces AGI before Jan 1, 2028", "ok": true, "message": "49.5% (0 backfill days)" }, { "source": "Polymarket \u00b7 OpenAI announces it has achieved AGI before 2027", "ok": true, "message": "8.5% (0 backfill days)" }, { "source": "Manifold \u00b7 Will we get AGI before 2030?", "ok": true, "message": "50.0% (0 backfill days)" }, { "source": "Kalshi \u00b7 Any company announces AGI before Jan 1, 2030", "ok": true, "message": "81.5% (0 backfill days)" }, { "source": "Manifold \u00b7 Massive catastrophe caused by AI before 2030?", "ok": true, "message": "39.9% (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": "112 incidents (generated 2026-10-20T18:00:19.633Z)" }, { "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)" } ] }