P(doom) interactive chart

Published on October 25, 2026 9:09 AM GMT TL;DR: I compiled 43 published estimates of AI catastrophic and existential risk into an interactive page: Forecasts of AI Loss-of-Control & Existential Risk. Summary: Inspired by Tom Cunningham’s compilation of forecasts of AI & economic growth, I asked Claude to help compile forecasts of catastrophic risk from AI. The page includes estimates from AI experts and researchers (28), AI lab founders/CEOs (4, from 3 people), superforecasters (4) and others (7: other forecasters, philosophers, commentators). The compilation is not exclusive - I'm sure I've missed out plenty of estimates. Overall: the median across all 43 is about 10% (IQR roughly 5–28%). The mean is 22%, pulled up by high estimates such as Yudkowsky (99.5%), Kokotajlo (~70%) and Conjecture staff (70–80%). The estimates aren’t directly comparable. By group: superforecasters sit roughly 10× below AI experts at matched horizons (median 0.4% vs 5% by 2100), though the gap narrows in the 2026 LEAP waves. Lab founders are spread widely (Altman ~2%, Amodei 25%). Caveats Forecasts vary substantially in terms of what counts as AI risk (e.g. takeover vs extinction vs catastrophe or unspecified 'very bad outcome') and  in their time horizon (from 10 years from AGI, to 2100, and a lot of unspecified). Some people provide several different estimates, and some estimates are secondary reports. The sample is obviously weighed towards people who spend a lot of time thinking about AI risk. It is worth noting that non-specialist supeforecasters generally have lower (though not necessarily more/less accurate) estimates. This is a compilation rather than an expression of my views, but any views expressed are my personal views and not those of my employer. Happy to receive feedback, and please let me know if you found this useful so I know whether its worth updating in the future.  Discuss