PromptZone - AI Prompts, Guides and Tools for Builders

Elina Watanabe
Elina Watanabe

Posted on

What Is P(doom) and Why AI Researchers Track It

A recent Hacker News thread on the term P(doom) drew 51 points and 23 comments. The discussion centers on how AI labs and independent forecasters assign a single probability to catastrophic AI outcomes.

What P(doom) Measures

P(doom) is shorthand for the probability that advanced AI systems cause human extinction or permanent disempowerment. The number is usually expressed as a percentage or decimal between 0 and 1.

Researchers derive the figure by breaking down failure modes: loss of control, misuse by bad actors, or gradual erosion of human agency. Each path receives its own conditional probability before the values are combined.

How Forecasters Arrive at the Number

Most public estimates come from surveys of AI researchers and dedicated forecasting platforms. Common inputs include:

  • Timelines to AGI
  • Alignment difficulty
  • Probability of successful containment after deployment

The HN thread highlighted that these inputs are rarely disclosed in full, making direct comparisons difficult.

Numbers Circulating in 2025–2026

Public estimates range from 1 % to 20 %. A 2023 survey of machine-learning researchers produced a median of 5 %. More recent statements from lab leaders sit between 10 % and 15 %.

Source Year Median P(doom)
ML researcher survey 2023 5 %
Public statements by lab CEOs 2025 10–15 %
Independent forecasters 2026 8–12 %

Community Reaction on Hacker News

Commenters focused on three points:

  • Lack of transparent models behind the percentages
  • Risk that a single headline number oversimplifies policy choices
  • Value of tracking P(doom) as one signal among many rather than the sole decision metric

Who Should Pay Attention

Safety teams at frontier labs use P(doom) to prioritize research. Policy analysts reference it when drafting compute thresholds or licensing rules. Individual developers and smaller companies gain little from adjusting their own work based on the number alone.

Practical Next Steps

Read the original post at the provided URL. Cross-check against the latest AI Impacts survey and Metaculus questions tagged “AI catastrophe.” Track updates rather than treating any single figure as fixed.

Bottom line: P(doom) remains a coarse but widely referenced shorthand for tracking existential risk estimates across labs and forecasters.

The metric will stay relevant only if the underlying models and assumptions become more transparent in the next round of surveys.

Top comments (0)