What is p(doom) in AI? Expert estimates

What the evidence says

The term began as informal shorthand in online rationalist circles. It was first used in a 2009 LessWrong comment and became widely used in 2023, after the release of GPT-4, when prominent researchers such as Geoffrey Hinton and Yoshua Bengio began warning about AI risks.1 In May 2023, the Center for AI Safety published a one-sentence statement: “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”11

Estimates people have stated in public, each linked to where they said it:

Geoffrey Hinton, AI researcher: a 10% to 20% chance that AI leads to human extinction within the next 30 years (BBC Radio 4, December 2024).5

Elon Musk: “I think there’s some chance that it will end humanity. I probably agree with Geoff Hinton that it’s about 10% or 20% or something like that” (Abundance Summit, March 2024).6

Dario Amodei, chief executive of Anthropic: a 25% chance that things go “really, really badly” with AI, and a 75% chance they go well (September 2025).7

Paul Christiano, AI alignment researcher: a 22% probability of an AI takeover, and 46% that humanity has “irreversibly messed up our future within 10 years of building powerful AI” (April 2023). He calls these rough guesses that vary from day to day.8

Joseph Carlsmith, philosopher: in his report “Is Power-Seeking AI an Existential Risk?” he estimated about a 5% chance of an existential catastrophe from power-seeking AI by 2070, and has since raised it to more than 10%.9

Surveys give a wider view. In a 2022 survey of machine learning researchers, the median respondent put 5% on the long-run effect of advanced AI being “extremely bad (e.g. human extinction)”.3 In a 2023 survey of 2,778 researchers who publish at top AI venues, between 38% and 51% gave at least a 10% chance to advanced AI leading to outcomes as bad as human extinction.2

A forecasting tournament found a large gap. In the Forecasting Research Institute’s 2022 Existential Risk Persuasion Tournament, the median superforecaster put the chance of AI-caused human extinction by 2100 at 0.38%, while the median AI expert put it at 3%; the superforecasters were largely unmoved by the experts’ higher estimates.4

Why the spread? The 2025 International AI Safety Report says expert opinion on the likelihood of losing control of AI “varies greatly: some consider it implausible, some consider it likely to occur, and some see it as a modest-likelihood risk that warrants attention due to its high potential severity.”10 For what a loss of control could look like, see our guide to AI takeover scenarios.