Forecasts on AI catastrophe, misuse, and governance from Metaculus, Manifold, Polymarket and Kalshi, next to published expert estimates. Each figure links to its source. Prices refresh every six hours from each platform's public API.
Weekly readings back to each question's opening, then one reading per day.
Metaculus implied figures come from multiplying its conditional questions. The Manifold line is a play-money extinction market, which is biased low.
Biggest moves in risk questions over the last 30 days, in percentage points. Select any question below to see its full history.
Probability of YES. Bars share one 0–100% scale. The change column compares today with 30 days ago.
Stated views, not markets. Updated by hand when new estimates are published.
Log scale. Bars show a stated range. Each person defines "doom" and the time horizon differently.
Structured, aggregated estimates.
Metaculus shows a community forecast with no money at stake. Manifold uses play money. Polymarket and Kalshi trade real money, so their AI questions lean short-term and policy-focused. The participation column shows forecasters, traders or dollar volume.
A market on human extinction can never pay out on YES, so traders have little reason to bid it up. Read the Manifold "wipe out humanity" prices as a lower bound on what traders believe, not a fair estimate.
Metaculus asks about AI catastrophe as a chain of conditionals. Rows marked "Implied" multiply the chain: P(10% population decline) × P(caused by AI | decline) × P(95% or more | AI catastrophe). Those rows are our calculation, not a Metaculus forecast.
Polymarket and Kalshi list no market on AI killing a share of humanity. Their AI questions cover regulation, lab behavior and AGI announcements. A market on AI killing 10% of people could still pay out to survivors, so its absence reflects listing and regulatory choices rather than the payout problem that breaks extinction markets.
"Catastrophe", "wipe out" and "p(doom)" mean different things on each platform and to each person. Compare rows only when their wording and time horizon match. Low-participation markets are shown but should carry little weight.