24 Oct 2026
Ranking v1
Ranking anchored on Ord (2020), XPT (2023) and AI Impacts (2023). Next: add Ord's 2024 Precipice Revisited commentary and any newer forecasting-tournament results.
← Research themes · Last updated 24 Oct 2026
Ranking the ways AI could end humanity
A sober ranking of high-level scenarios, anchored to published estimates.
Working rankingOf the ways advanced AI could end or permanently cripple humanity, loss of control of misaligned systems ranks highest in published estimates. Misuse (especially biology) and slow, systemic disempowerment follow, but the uncertainty is enormous.
High-level mechanisms only; there is no operational detail on this page. Published probabilities are subjective expert or forecaster judgements and differ by more than an order of magnitude. Each source's own number is shown with its definition (extinction vs. catastrophe, time horizon).
Scenario ranking
7 rows
| # | Scenario | Mechanism (high level) | Published estimates (source's own numbers) | Basis |
|---|---|---|---|---|
| 1 | Loss of control of misaligned advanced AI | Highly capable systems pursue goals that conflict with human interests, and humans lose the ability to correct or stop them. | Ord 2020: 1 in 10 (unaligned AI, existential catastrophe within the next 100 years, as estimated in 2020). XPT: AI-caused extinction by 2100, 0.38% (superforecasters) vs 3% (experts). AI Impacts 2023: median 5% for extinction or severe disempowerment; median 10% for loss of control causing extinction. | Sourced |
| 2 | AI-assisted engineered pandemic (misuse) | AI lowers barriers for a malicious actor to cause a biological catastrophe. | Ord 2020: 1 in 30 for engineered pandemics (all causes, not AI-specific). XPT: engineered-pathogen extinction by 2100, 0.01% vs 1%. AI Impacts 2023: 73% of respondents rated engineered viruses from dangerous groups a substantial or extreme concern. A RAND red team (2024) found no measurable uplift from the LLMs it tested. | Sourced (not AI-specific) |
| 3 | Gradual disempowerment | Incremental handover of economic, political and cultural functions to AI erodes human influence without any single catastrophic event. | No published probability. Kulveit et al. 2025 argue it is a systemic existential risk. | None published |
| 4 | AI-enabled power concentration or authoritarian lock-in | A small group uses AI to seize or entrench power irreversibly (an "unrecoverable dystopia"). | No probability. AI Impacts 2023: 73% rated authoritarian control a substantial or extreme concern. See Forethought. | None published |
| 5 | AI-accelerated great-power or nuclear war | AI races, compressed decision times or destabilised deterrence raise the odds of a nuclear exchange. | Ord 2020: 1 in 1,000 for nuclear war (all causes). XPT: nuclear-caused extinction by 2100, 0.074% vs 0.55%. No AI-specific figure. | Sourced (not AI-specific) |
| 6 | Attacks on critical infrastructure | AI-scaled cyber operations against power, water or finance cause cascading failures. | No probability. Evidence of trend: Anthropic (2025) reported a largely AI-orchestrated espionage campaign; CISA AA24-038A documented state actors pre-positioned in US infrastructure. | None published |
| 7 | Epistemic collapse | Deepfakes and mass persuasion degrade a society's ability to coordinate a response to any other risk. It amplifies other risks rather than causing extinction directly. | No probability. AI Impacts 2023: 86% rated deepfakes and 79% rated manipulation of public opinion a substantial or extreme concern. | None published |
Sort by any column. "Sourced (not AI-specific)" means the number covers the hazard from all causes, so AI's contribution is only part of it.
How the ranking was formed
- Scenarios with a published probability come first, ordered by the source's own central number: Ord's per-century estimates (1/10 > 1/30 > 1/1,000), cross-checked against the XPT ordering (AI > nuclear > engineered pathogens for extinction among experts).
- Scenarios without any published probability come next, ordered by how many independent sources treat them as a path to existential (not just catastrophic) harm, then by the AI Impacts concern share. This step is a judgement call and is marked "None published".
- Gradual disempowerment and power concentration are placed above nuclear war because the sources treat them as AI-specific existential paths, while the nuclear numbers are mostly not about AI. Reasonable people would order these differently.
What the totals look like. XPT medians for all-cause extinction by 2100 were 1% (superforecasters) and 6% (experts). Ord put total existential risk this century at about 1 in 6.
What to watch
- Updated forecasts: an XPT follow-up, the next AI Impacts survey, and Ord's 2024 update (The Precipice Revisited).
- Frontier-lab safety-level escalations (e.g. Anthropic ASL-3 activation, May 2025).
- Evidence of scheming or deception in evaluations (see Theme 11).
Reading
- Existential Risk Persuasion Tournament— Forecasting Research Institute, 2023Superforecaster vs. expert probabilities.
- Thousands of AI Authors on the Future of AI— Grace et al. (arXiv), 2024-01Survey of 2,778 researchers.
- On the Extinction Risk from Artificial Intelligence— RAND, 2025-05-06Finds extinction "immensely challenging" but not ruled out. A sceptical anchor.
- An Overview of Catastrophic AI Risks— Hendrycks, Mazeika & Woodside, 2023A taxonomy: malicious use, AI race, organisational risks, rogue AIs.
- Is Power-Seeking AI an Existential Risk?— Carlsmith, 2022A step-by-step probability argument.
- International AI Safety Report— International AI Safety Report, 2026A government-backed scientific consensus review.
Signals
Signals the risk is rising
- Labs raising safeguard levels for CBRN misuse (Anthropic ASL-3).
- Evaluations finding in-context scheming (Apollo 2024) and agentic misalignment (Anthropic 2025).
- Task horizons of AI agents doubling quickly (METR; see Theme 11).
Signals it is lower or manageable
- Red-team studies finding little uplift (RAND 2024).
- RAND (2025): extinction would require AI to have several capabilities at once and would unfold slowly, giving time to respond.
- Superforecasters' far lower numbers (XPT).
What to watch
- The signed Statement on AI Risk, and whether new signatories or retractions appear.
Reading
- Activating ASL-3 protections— Anthropic, 2025-05A frontier lab escalating safeguards.
- Agentic Misalignment— Anthropic, 2025-06Stress tests of model behaviour.
- Statement on AI Risk— Center for AI Safety, 2023-05A one-line statement signed by many leading researchers.
Prediction markets
Live prices from public prediction markets, snapshotted by scripts/update_markets.py. A price of 30% means traders pay about 30¢ for a contract that pays $1 if the event happens; treat it as a crowd estimate, not a fact. Polymarket and Kalshi are real-money markets (Kalshi is CFTC-regulated); Manifold uses play money and is often thin. Long-dated and low-volume markets are poorly calibrated. Each market's exact resolution rules on the platform decide what counts, so read them before relying on a headline question.
Superintelligence timing
36%
Ṁ82k play-money · 243 traders · closes 2030-01-01 · as of 2026-10-30 17:06 UTC
75%
Ṁ96k play-money · 273 traders · closes 2040-01-01 · as of 2026-10-30 17:06 UTC
Manifold Will we get AGI before 2030?
49%
Ṁ140k play-money · 347 traders · closes 2030-03-01 · as of 2026-10-30 17:06 UTC
6% bid 6¢ / ask 7¢
367k contracts · 175k open interest · closes 2027-01-01 · as of 2026-10-30 17:06 UTC
41% bid 40¢ / ask 43¢
301k contracts · 137k open interest · closes 2028-01-01 · as of 2026-10-30 17:06 UTC
64% bid 62¢ / ask 66¢
209k contracts · 84k open interest · closes 2030-01-01 · as of 2026-10-30 17:06 UTC
Catastrophe markets (long horizon, play-money)
3%
Ṁ106k play-money · 139 traders · closes 2030-01-01 · as of 2026-10-30 17:06 UTC
8%
Ṁ24k play-money · 71 traders · closes 2039-12-31 · as of 2026-10-30 17:06 UTC
44%
Ṁ9k play-money · 56 traders · closes 2100-09-09 · as of 2026-10-30 17:06 UTC
Manifold Will there be a botched AI takeover/attempted shutdown evasion killing >100 people before 2050?
27%
Ṁ3k play-money · 54 traders · closes 2050-01-02 · as of 2026-10-30 17:06 UTC
3%
Ṁ27k play-money · 72 traders · closes 2030-01-01 · as of 2026-10-30 17:06 UTC
Pandemic and safety governance
Polymarket New pandemic in 2026?
3%
$1.3M vol · $40k liquidity · closes 2027-01-01 · as of 2026-10-30 17:06 UTC
Polymarket New pandemic before 2028?
14%
$29k vol · $5k liquidity · closes 2028-01-01 · as of 2026-10-30 17:06 UTC
8% bid 4¢ / ask 8¢
15k contracts · 11k open interest · closes 2027-01-01 · as of 2026-10-30 17:06 UTC
23% bid 24¢ / ask 32¢
139 contracts · 66 open interest · closes 2027-07-01 · as of 2026-10-30 17:06 UTC
48% bid 43¢ / ask 48¢
101 contracts · 28 open interest · closes 2028-01-01 · as of 2026-10-30 17:06 UTC
18%
$9k vol · $520 liquidity · closes 2027-01-01 · as of 2026-10-30 17:06 UTC
Polymarket U.S. enacts AI safety bill before 2027?
9%
$207k vol · $19k liquidity · closes 2027-01-01 · as of 2026-10-30 17:06 UTC
Snapshot fetched 2026-10-30 17:06 UTC from public APIs (Polymarket Gamma/CLOB, Kalshi trade API, Manifold v0). Prices are what traders are paying, not my forecasts and not advice; thin or long-dated markets can be far off. Sparklines show daily prices over up to 120 days. Metaculus: API now requires an authenticated account token (HTTP 403 without one) and the website is behind a bot challenge, so Metaculus forecasts are not included. Add a ref type with a personal token later if wanted.
How to add a market
Add a ref to "markets" → "groups" → "refs" in src/themes/<slug>.json: {"platform":"polymarket","event":"<slug after /event/ in the URL>"}, {"platform":"kalshi","event":"<EVENT_TICKER>"} or {"platform":"manifold","id":"<market id or slug>"}. Optional: "markets" (list of specific sub-market slugs or tickers), "max", "sort" (date, prob or api). Then run python3 scripts/update_markets.py <slug> and rebuild.
Notes & journal
Dated observations from src/notes/ai-extinction-scenarios.md. Add a new entry with a ## YYYY-MM-DD Title heading.