US unemployment exceeds 25%

Where things stand

• Latest reading. The latest official reading of LNS14000000 is 4.2% for September 2026, published 2 Oct 2026. The BLS API returns Jul 4.1, Aug 4.1, Sep 4.2. That month is eligible but nowhere near the threshold. A YES needs a rise of about 21 percentage points.
• Upcoming releases. BLS's schedule (https://www.bls.gov/schedule/news_release/empsit.htm) puts October data on 6 Nov 2026 and November data on 4 Dec 2026. By end-2031 the last publishable month is November 2031.
• Current labour market. It is "low-hire, low-fire": 29k jobs added in September, AI is the most-cited reason in Challenger layoff announcements, and there is an Iran war and inflation stress (CNN, 2 Oct 2026; Challenger, Oct 2026). None of this is close to the scale needed. The October reference week is this week, so Q4 2026 is effectively already decided.

Base rate (computed)

• Data. I pulled all 944 monthly values from 1948 to Sep 2026 via the BLS API (api.bls.gov/publicAPI/v2/timeseries/data/LNS14000000). The series maximum is 14.8% (April 2020), and it has never reached 15%. The largest rise inside any five-year window was +11.3 points (Sep 2019 3.5% → Apr 2020 14.8%). Only the 1933 Great Depression year (about 24.9% annual, from historical estimates rather than this survey) comes near 25%.
• Tail fits. I fitted the trough-to-peak rises of 10 post-war cycles.
  • A heavy-tailed Pareto fit (tail index about 1.9, driven by COVID) gives roughly 1.4% per recession for a 21-point rise. At about 0.6 recessions per five years, that is about 0.9%.
  • An exponential fit gives about 0.01%.
• So the history-based prior runs from about 0.01% to 1%.

Forecasts from other sources

• Expert surveys and models.
  • The Forecasting Research Institute survey (March 2026, https://forecastingresearch.org/research/economic-effects-of-ai) has a median 2030 unemployment forecast of about 5%. Its rapid-AI scenario has a median of 6% and a 90th percentile of about 10–11%.
  • Anthropic's Econ Scenario Explorer (Sept 2026) puts its extreme scenario at about 12% overall unemployment in 2030.
  • A National Bureau of Economic Research paper (Dynan, Elmendorf and Sheiner) has about 2% of the labour force displaced at a time.
• Markets and forecasting sites.
  • Kalshi prices U-3 above 10% in December 2030 at about 9–10%, a thin long-dated market.
  • Metaculus gives about 55% for 10% before 2031, and 5% for the employment-to-population ratio falling below its COVID low (51.3%) in 2030.
  • Manifold markets on 25% or more show 9–17%. They have about 25–30 traders, price 20% and 25% almost the same, and long-dated tail markets there are biased upward, so I discounted them heavily.

Paths to YES

1. A pandemic with broad shutdowns. COVID's April 2020 rate would have been about 19.7% without the misclassification BLS reported (BLS, 8 May 2020), so a shock about 1.3 times as large could cross 25%.
2. AI-driven mass displacement in a fast takeoff. This would need tens of millions displaced quickly, little re-employment, and displaced people still looking for work.
3. Other catastrophes, such as grid failure, war or financial collapse, where BLS still keeps publishing.

Paths to NO

• Displaced workers tend to leave the labour force, which keeps the headline rate down.
• Large policy responses would follow: transfers, job programmes, payroll support.
• Physical bottlenecks and slow adoption limit how fast AI can replace work.
• In the worst catastrophes BLS might not publish at all.

Hazard model

• Non-AI shocks: about 0.025% per quarter, lower in the first two quarters because no shock is under way.
• AI component: near zero in 2027, rising to about 0.05% per quarter by 2031.
• Totals: about 0.5% from non-AI shocks and 0.46% from AI, so cumulative probability is about 0.1% by end-2027, 0.43% by end-2029 and 0.95% by end-2031.

Strongest case against my estimate

The Manifold markets and the high Metaculus 10% forecast suggest some forecasters see much fatter AI tails. A true intelligence explosion around 2027–28 with gridlocked policy could push the rate past 25%. I let this shape the rising AI hazard. It did not move me further, because even the authors of the extreme scenarios put unemployment at roughly 12–20%.

Question. By the last day of each quarter, has BLS published a US unemployment rate exceeding 25% for any month?

Definitions.

• US unemployment rate: the monthly, seasonally adjusted U-3 unemployment rate from the Bureau of Labor Statistics (BLS) Current Population Survey, series LNS14000000. This is total unemployed as a percentage of the civilian labor force aged 16 and over in the 50 states and District of Columbia, excluding institutional residents and active-duty Armed Forces personnel.
• Exceeding 25%: the published rate is strictly greater than 25.0%; at the current one-decimal precision, 25.1% or higher qualifies. Use the published rate, not a ratio independently calculated from employment counts.
• Eligible month: September 2026 or later, under the Observation period convention.

Resolves YES if an initial release or an official BLS revision publishes a qualifying rate for at least one eligible month by the quarter-end. No minimum duration or attribution to AI is required. A revision that first raises a rate above the threshold qualifies on its publication date; a later downward revision does not undo YES under the Resolution window convention.

Resolves NO if no qualifying publication exists by the quarter-end. A rate of exactly 25.0%, unadjusted rates, U-6 or other alternative measures, state or demographic-subgroup rates, forecasts, unofficial estimates, and revisions solely to months before September 2026 do not qualify. Missing or delayed observations do not themselves qualify.

Resolving source. BLS only: The Employment Situation, Table A-1, TOTAL, seasonally adjusted unemployment-rate row; Table A-15, seasonally adjusted U-3 row; or the official LNS14000000 series and its dated BLS revisions. A BLS-designated successor series counts only if it retains the national, monthly, seasonally adjusted U-3 measure and population coverage. Apply the Resolution window, Observation period, and Resolution date (A3) conventions; the relevant date is publication, not the reference month.