p(doom)
Definition
p(doom) denotes the subjective probability that someone assigns to an existentially catastrophic outcome from AI for humanity.
p(doom) is a subjective probability that artificial intelligence will have existentially catastrophic consequences for humanity. What counts as “doom” varies; usually extinction, permanent disempowerment of humanity, or a collapse of civilization are meant. There is no standardized calculation method.
Simply explained
Like a personal betting odds: it shows an assessment but is not a measurement.
Why it matters in practice
Helps to contextualize public debates about AI risks: researchers, policymakers, and industry representatives use the term to express their assessment of existential AI risks, even though there is no standardized method. For concrete corporate projects, the number does not replace a dedicated risk assessment.
Example from business
Possible example (hypothetical): In an AI strategy training, a team explains why publicly cited p(doom) values are assessments and do not describe project risks.
Limits and pitfalls
- Treating personal estimates as measurements; comparing values with different definitions of “doom”; not clarifying which assumptions (e.g., about development timelines, alignment difficulty, or governance) underlie a value; mixing existential risks with project risks.
Origin & history
p(doom) was described in 2025 as an expression of subjective assessments of existential AI risks. A related survey of 2,778 AI researchers took place in 2023 and was published in 2024; it did not use the term itself.
Use in mid-sized companies
- Possible applications: contextualizing media reports in AI trainings, discussing AI safety in strategy workshops, distinguishing from operational AI risk management.
Frequently asked questions
Is p(doom) an objective measurement?
No, a standardized calculation is lacking; the value expresses personal assumptions.
Does it describe the risk of a corporate chatbot?
No, it refers to existential consequences for humanity.
What does the survey by Grace and others show?
Depending on the question, 37.8–51.4 percent estimated the risk of extinction-like outcomes at at least ten percent. The authors emphasize that expert forecasts do not guarantee objective truth.
Sources
- The AI Risk Spectrum: From Dangerous Capabilities to Existential Threats · Grey & Segerie (2025): The AI Risk Spectrum: From Dangerous Capabilities to Existential Threats
- Thousands of AI Authors on the Future of AI · Grace et al. (2025): Thousands of AI Authors on the Future of AI
Sources checked on 14.10.2026. Editorial use cases are hypothetical; no evidence of real customer results.
Editorially reviewed on: