US civilian unemployment rate above 10% for 12 consecutive months

---
title: US civilian unemployment rate above 10% for 12 consecutive months
status: draft
dimensions: ["labor","finance","governance"]
horizon: medium
trigger: US civilian unemployment rate (BLS series LNS14000000, as charted at https://www.bls.gov/charts/employment-situation/civilian-unemployment-rate.htm) is above 10.0% for 12 consecutive monthly readings, seasonally adjusted.
timeline: {"p10":2030,"p50":2038,"p90":2070}
confidence: low
sub_gates: [{"slug":"us-recession-2026-2028","p50":2027,"why":"A cyclical recession in the 2026-2028 window is the most plausible bridge from today's 4.3% unemployment to the 10% threshold. NY Fed term-spread model is at ~19% by Jan 2027; Polymarket/Kalshi pricing 16% for 2026, 45% for 2027. A standard recession alone won't sustain 12-month >10% (only the Great Depression did), but it sets the level from which an AI-driven structural overlay would have to push further."},{"slug":"us-unemployment-touches-10","p50":2031,"why":"A single-month print of >=10% — much easier than 12 consecutive months. Only happened in 1982-83 (10 months above 10%, peak 10.8%) and Oct 2009 (10.0%, one month) and Apr 2020 (14.7%, two months) post-WWII. Metaculus 'US unemployment rate 10% before 2031' trades at ~1%. I'd put a single touch around 2031 median, dominated by either AI displacement OR a deep recession OR both compounding."},{"slug":"ai-displaces-5pct-workforce-net","p50":2032,"why":"Net displacement (gross AI displacement minus AI-enabled new roles minus reabsorption into other sectors) of 5%+ of the US labor force — about 8M jobs net. CFO surveys (NBER/Duke) project 502K AI-driven cuts in 2026 (~0.4% of workforce); Goldman base case is 6-7% gross displacement over a decade with 0.6pp net unemployment impact. Net 5% requires AI capability to outpace Baumol/reabsorption mechanisms — possible if METR doubling stays at 4 months and humanoids commercialize at scale by ~2030."},{"slug":"labor-reabsorption-fails","p50":2035,"why":"The Schumpeter/creative-destruction reabsorption mechanism that has reliably moved displaced US workers into new sectors since 1850 fails. Requires that AI agents AND humanoids substitute for both knowledge work AND physical labor faster than the economy can spin up new occupations. Historically, occupation churn takes 10-30 years; AI compresses the substitution side to <5 years. If the new-occupation creation side doesn't accelerate proportionally, you get persistent surplus labor."},{"slug":"policy-fails-to-bridge-12mo","p50":2036,"why":"Even if structural unemployment spikes, the US has many policy levers (fiscal stimulus, expanded UI, public works, tariffs) historically used to compress >10% episodes back below threshold within months. Bush/Obama stimulus 2009-10 and CARES 2020 are recent precedents. For 12 consecutive months above 10%, the policy response must either fail (political gridlock), arrive too late, or be insufficient to offset the structural shock — a high bar given the political costs of double-digit unemployment."}]
history: [{"date":"2026-05-31T00:00:00.000Z","p10":2030,"p50":2038,"p90":2070,"why":"Initial estimate from initial research."}]
cross_gate: [{"other":"ai-agent-30pct-knowledge-work","relation":"enabled_by","strength":"strong","note":"Most direct upstream gate. If AI agents handle >=30% of knowledge-work tasks by ~2029 (that gate's P50), gross displacement of US white-collar workers (estimated 60% of employment) hits ~18% of the labor force on a tasks-to-jobs translation. Net unemployment depends on (a) part-task-vs-whole-job substitution (Amodei's Jevons reframe), (b) speed of new-role creation, and (c) labor-force-participation response. A baseline 'AI handles 30% of tasks' world adds 1-3pp to US unemployment per Goldman/Acemoglu; getting from there to a sustained 10%+ requires another shock or much faster substitution."},{"other":"humanoid-retail-20k","relation":"enabled_by","strength":"medium","note":"Once humanoids hit $20K commercial price (that gate's P50 = 2028), the substitutability frontier extends from knowledge work into physical service work — retail clerks, warehouse pickers, food prep, basic facility maintenance. This is the channel by which mass unemployment scales beyond white-collar (~60% of US jobs) into the broader workforce. Without it, AI-driven unemployment plateaus around the knowledge-work fraction."},{"other":"humanoid-10m-households","relation":"enabled_by","strength":"medium","note":"10M household humanoids (P50 2035) implies fleet-scale humanoid manufacturing economics, which means commercial deployment in business contexts is far cheaper. This is the demand-side enabler for humanoid-retail-20k and therefore for the physical-labor channel into mass unemployment."},{"other":"humanoid-self-replication-factory","relation":"enabled_by","strength":"medium","note":"If humanoid factories self-replicate (P50 2034), the supply curve for embodied labor becomes effectively unbounded at marginal cost ~= materials+energy. This is the singular technological event that would most plausibly drive sustained 10%+ structural unemployment, because the wage floor for substitutable human labor collapses to subsistence."},{"other":"global-economy-explosive-growth","relation":"substitutes","strength":"strong","note":"Counter-intuitive but important. If global GDP doubles by 2049 (that gate's P50), the economy is producing so much output that labor demand for non-substitutable work (creative, interpersonal, novel-context, embodied-but-not-yet-roboticizable) explodes. Sustained 10%+ unemployment in a doubled-GDP world requires that essentially all human capabilities become AI-substitutable — a higher bar than just AI handling knowledge work. The two gates can both fire but the explosive-growth path makes 12-month-sustained-10% LESS likely, not more. Conversely, the AI capability surge can fire WITHOUT the GDP boom landing (a 'productivity-but-no-growth' scenario) — this is the worst-case world for unemployment."},{"other":"ai-tutor-k8-parity-20mo","relation":"substitutes","strength":"weak","note":"AI tutors at K-8 parity (P50 2031) is the leading edge of AI-driven reskilling capacity. If displaced workers can be rapidly retrained via AI tutors into still-human roles, the reabsorption mechanism strengthens and 12-month-sustained-10% becomes less likely. Effect is real but small relative to the speed-of-substitution channel."},{"other":"human-aging-halted","relation":"correlates","strength":"weak","note":"Both are 'transformative AI' downstream gates — both fire in worlds with very capable AI. If aging is halted by 2048, working lifetimes extend dramatically, increasing labor supply and creating a 'gerontocracy unemployment' angle that compounds the AI-substitution channel. Distant interaction; second-order effect."},{"other":"corporate-sovereignty-territory","relation":"correlates","strength":"weak","note":"Sustained 10%+ unemployment is one of the political-economy conditions under which corporate sovereignty becomes thinkable — large populations dependent on private-sector wealth redistribution rather than state employment. Plausible but speculative chain; same direction, distant nodes."},{"other":"autonomous-resource-frontier-positive-roi","relation":"correlates","strength":"weak","note":"If autonomous mining/space resource extraction goes positive ROI, capital flows further away from labor-intensive activities. Marginal effect on US unemployment, but in the same family of capital-replacing-labor dynamics."},{"other":"autonomous-freight-delivery","relation":"enables","strength":"medium","note":"Trucking + last-mile delivery employs 3.5M Americans (~2% of US workforce). Full automation of long-haul + last-mile would directly add ~1.5pp to unemployment within the displacement window, before reabsorption. One of the most-direct labor-channel gates."},{"other":"robotaxi-unit-economics-5-cities","relation":"enables","strength":"weak","note":"Robotaxi at-scale displaces ~1.5M US drivers (rideshare + taxi). Material but small fraction of the path to 10%."},{"other":"construction-robot-40pct-labor","relation":"enables","strength":"weak","note":"Construction employs ~8M Americans. 40% labor reduction frees up ~3M, contributing to the displacement count. Modest gate-specific contribution."}]
key_dependencies: [{"factor":"AI agent knowledge-work displacement rate","kind":"gate","direction":"accelerates","linked_gate":"ai-agent-30pct-knowledge-work","impact":"If AI agents handle >=30% of knowledge-work tasks by ~2028, gross white-collar displacement hits ~18% of the labor force on a tasks-to-jobs translation, the primary structural mechanism driving unemployment toward 10%."},{"factor":"Humanoid robot commercial price point","kind":"gate","direction":"accelerates","linked_gate":"humanoid-retail-20k","impact":"Once humanoids reach $20K commercial price (~2028 P50), substitutability extends from white-collar into physical service work — retail, warehouse, food prep — scaling displacement well beyond the ~60% knowledge-work fraction and making sustained 10%+ materially more reachable."},{"factor":"Humanoid self-replication manufacturing","kind":"gate","direction":"accelerates","linked_gate":"humanoid-self-replication-factory","impact":"Self-replicating humanoid factories make embodied labor supply effectively unbounded at marginal cost; this singular event would most plausibly drive sustained structural unemployment above 10% by collapsing the wage floor for substitutable human labor."},{"factor":"Explosive global GDP growth","kind":"gate","direction":"delays","linked_gate":"global-economy-explosive-growth","impact":"If global GDP doubles by 2049, labor demand for non-substitutable work explodes, making sustained 10%+ unemployment less likely conditional on AI capability gates firing; worst-case for unemployment is AI productivity surge without the GDP boom landing."},{"factor":"AI capability doubling rate sustaining","kind":"data","direction":"accelerates","linked_gate":null,"impact":"METR's Jan 2026 update shows task-completion time horizon doubling every 4.3 months; if sustained this implies ~10x agent capability every 14 months, directly compressing the displacement timeline by years relative to the median economist baseline."},{"factor":"Federal policy response to unemployment spike","kind":"regulation","direction":"delays","linked_gate":null,"impact":"Trump EO 14365 limits AI regulation and the administration is hostile to UI expansion; if a displacement shock arrives in 2026-28, a slower and smaller federal response than 2009 ARRA or 2020 CARES materially raises the probability a >10% episode sustains for 12+ months rather than being compressed below threshold."},{"factor":"Labor force participation collapse masking U-3","kind":"data","direction":"both","linked_gate":null,"impact":"If displaced workers exit the labor force rather than register as unemployed (participation already at multi-year low of 61.8%), the official U-3 rate can stay below 10% even as the economic dislocation is Great-Depression-class, meaning the gate trigger is harder to reach despite severe underlying conditions."}]
external_calibration: {"metaculus":"https://www.metaculus.com/questions/18664/us-unemployment-rate-10-before-2031/","manifold":"https://manifold.markets/ahalekelly/will-ai-cause-the-us-unemployment-r","expert_consensus":"Sharp disagreement. Dario Amodei (Anthropic CEO) said in May 2025 that AI could push US unemployment to 10-20% within 5 years as half of entry-level white-collar jobs disappear — then walked the claim back in May 2026, reframing via Jevons paradox. Goldman Sachs base case: 6-7% gross displacement over a decade, ~0.6pp net unemployment impact, which would leave US unemployment in the 5-6% range, not 10%. Daron Acemoglu (Nobel laureate, skeptic): AI lifts TFP by 0.66% over a decade, not the 7-10% boom others forecast — implies sub-1pp unemployment impact. Daniel Kokotajlo (AI 2027): mass displacement starts late 2026, median timeline for AI automating >95% of remote jobs slipped from 2028 to 2029-30 in his 2026 updates; doesn't forecast unemployment rate directly but the scenario implies very rapid displacement. Fed/CBO baseline: unemployment near 4.4% through 2030. The 'sustained 12-month >10%' criterion is unprecedented in post-WWII US history (1982-83 was 10 months, Great Recession was 1 month at 10%, COVID was 2 months at >10%) — only the Great Depression met this bar. Metaculus 'US unemployment 10% before 2031' trades at ~1%; Manifold 'AI cause US unemployment >10% before 2030' single digits; Kalshi traders price 47% odds unemployment above 9% by 2030."}
last_updated: "2026-05-31T00:00:00.000Z"
sources_count: 76
---

## TL;DR

I put the **P50 at 2038** — about 12 years out from today (May 2026) — that the US civilian unemployment rate stays above 10% for 12 consecutive months, with a wide band: P10 = 2030 (a near-term recession plus rapid AI substitution arriving roughly together), P90 = 2070 (the gate may never fire on any plausible policy-responsive trajectory; I'm using 2070 as a "doesn't fire this century" anchor). **Confidence: low** — this is one of the gates where the answer depends almost entirely on the speed of AI labor substitution vs. the speed of new-role creation and policy response, and that ratio is the single most-debated parameter in 2026 economic forecasting.

The headline framing: **sustained 10%+ unemployment for a year has only happened once in modern US history — during the Great Depression**. The 1981-82 Volcker recession hit 10 months above 10% (peak 10.8% Nov 1982); the 2009 Great Recession touched 10.0% for one month (Oct 2009); COVID spiked to 14.7% in April 2020 but was back below 10% by August. The Reagan/Volcker episode is the closest post-WWII analog and it still missed the 12-month bar by 2 months. So the gate as written is a stricter criterion than most casual readers parse — it requires *either* a recession deeper than 1981-82 *or* a structural shock with no policy compensation, sustained.

The bull case (P10 = 2030): a cyclical 2026-28 recession (currently priced ~45% by 2027 on Kalshi) lifts unemployment to 7-8% in the standard way, AI-driven structural displacement (already visible: 142K tech layoffs in 2026 H1, 502K projected by NBER/Duke CFO survey) adds 2-3pp on top, and labor-force participation collapse + policy gridlock (Trump admin opposes UI expansion and rolled back state AI rules per EO 14365) prevents the usual compression. That stack reaches 10%+ by 2028-29 and could sustain through 2030. Note: this requires AI substitution moving 3-5x faster than Acemoglu's "0.66% TFP over a decade" baseline.

The bear case (P90 = 2070+, "never"): the modal economist (Acemoglu, CBO, Goldman base case) is right that AI adds <1pp to unemployment, reabsorption works the way it always has since 1850, and even a deep cyclical recession compresses back within 6-10 months under standard fiscal/monetary policy. In this world the gate trigger is a Great Depression-class shock or a Carrington-event-class infrastructure failure — neither of which is the AI-substitution thesis this gate is really about.

The reason I land at P50 = 2038 and not later: METR's January 2026 update has AI agent task-completion time horizon doubling every **4.3 months** (down from 7 months previously), 51% of AI 2027's predictions are confirmed/ahead-of-schedule, Anthropic Economic Index shows hiring of 22-25-year-olds in AI-exposed roles is already down 14% since ChatGPT, and CFO surveys project 502K AI-driven cuts in 2026 alone (9x the 2025 rate). These are early-warning indicators consistent with the displacement-faster-than-reabsorption thesis. They don't yet show up in the headline 4.3% unemployment rate, but the leading edges (U-6 at 8.2%, labor force participation at 61.8% — lowest since Oct 2021, entry-level grad unemployment at 9.7% — worst in 37 years) suggest the official rate is masking weakening.

The gate also implicitly bets on a question that economists are NOT in consensus about: **does AI substitution behave like prior automation waves (creative-destruction reabsorption works) or differently (substitution outpaces creation)?** I'm assigning meaningful (~35-50%) probability mass to "differently" by 2040, which is more bearish than Acemoglu but more conservative than Amodei's 2025 framing. Hence P50 = 2038, not 2032 (Amodei) and not 2080+ (Acemoglu).

## Current state (as of 2026-05-31)

**Headline number, May 2026 baseline**:
- **April 2026 US civilian unemployment rate: 4.3%**, unchanged from March 2026, little changed from April 2025 (BLS Employment Situation, released May 2026) [1]. May 2026 number releases June 5, 2026 — not yet available as of today.
- Twelve-month trend: 4.3% → 4.3% → 4.4% (Feb 2026) → 4.3% (Mar) → 4.3% (Apr). Gradually drifting up from the ~3.6% post-COVID lows of 2022-23 but still well below the post-WWII average of ~5.7% [2].
- **Distance to gate trigger**: would need to **more than double** from current 4.3% and stay there for 12 straight monthly prints.

**Beneath the headline — leading indicators of softening**:
- **U-6 (broader unemployment + discouraged + part-time-for-economic-reasons)**: 8.2% in April 2026, up from 8.0% in March 2026. Part-time-for-economic-reasons climbed 445K to 4.9M.
- **Labor force participation rate**: 61.8% in April 2026 — **lowest since October 2021**. Falling participation is masking some of the weakening (people exiting the labor force don't count as "unemployed" by the U-3 definition this gate uses).
- **Sahm Rule** (3-month MA unemployment minus 12-month low): 0.30 as of Feb 23, 2026. Trigger threshold is 0.50. Was triggered in July 2024 (0.54) but did not coincide with recession that year ("soft landing"). Currently elevated but not flashing.
- **Initial jobless claims**: 215K (week ended May 23, 2026), highest in over a month. Continuing claims 1.79M. Still firmly below 2024 averages but trending up week-over-week through May.
- **Entry-level grad unemployment**: 9.7% — **worst in 37 years**. About 43% of recent US grads are underemployed (working jobs not requiring a degree). Finance/info services were the historical on-ramp; those sectors were adding 44K jobs/month pre-pandemic and have been losing 9K/month since 2023.

**The AI-displacement signal in the 2026 data**:
- **Tech layoffs in 2026 H1: 142,000** (Tom's Hardware tracker), projected ~370K full-year. Meta, Amazon, Microsoft, Oracle named explicitly. Meta cut 8,000 in early May 2026 with Zuckerberg explicitly tying it to the AI capex budget (the company "chose to buy GPUs instead").
- **Big-4 hyperscaler 2026 capex**: ~$725B (77% YoY increase) per Invezz / 24/7 Wall St analysis — directly displacing tech labor and going to data centers + GPUs.
- **CFO survey (NBER/Duke/Fed Banks, 750 CFOs)**: AI-driven cuts projected at ~502,000 for 2026 (~0.4% of US workforce), **9x the 2025 reported number of 55K** (Challenger Gray & Christmas). 44% of CFOs plan some AI-related cuts.
- **Mercer 2026 Global Talent Trends**: 99% of CEOs say they expect AI-driven layoffs in next 2 years.
- **Anthropic Economic Index (Feb 2026 report, "Learning Curves")**: no clear spike in unemployment in most AI-exposed occupations yet, but **hiring of 22-25-year-olds into AI-exposed roles is down ~14% since ChatGPT launched**. Customer service, programmers, financial analysts are the top-3 AI-exposed.
- **Israeli analog (Jerusalem Post, Calcalist trackers)**: Israeli unemployed share from AI-exposed occupations went from 14-16% in 2019-22 to 20-25% in 2025. Israel is a leading indicator for the US because Israeli tech is structurally similar but smaller/faster-moving.

**Macro recession risk (the cyclical channel)**:
- **Polymarket "US recession by end of 2026"**: ~16% as of late May 2026.
- **Kalshi recession-by-2026**: ~16%; **2027**: ~45%; longer-horizon contracts price 60% chance US unemployment crosses 8% before 2030, 47% chance exceeding 9% before 2030, 75% chance unemployment is above 7% in 2030.
- **NY Fed term-spread model**: ~18.7% probability of recession by Jan 2027 (Feb 2026 reading).
- **Deloitte Q1 2026 baseline**: 2.0-2.3% real GDP growth for 2026, no recession, but rising-stagflation tail risk. Deloitte alternative scenario sees unemployment rising to 6.5% by 2028.
- **Fed FOMC March 2026 SEP**: median 2026 unemployment 4.4%, 2027 4.3%, 2028 4.2% — i.e., the Fed sees gradual normalization, no spike.

**Historical reference for the 12-month sustained 10% trigger**:
- **Great Depression**: unemployment above 14% from 1931 to 1940 — comfortably above the 12-month-at-10% bar for **~9 consecutive years**. The only US episode that clears this gate.
- **1981-82 Volcker recession**: unemployment above 10% from September 1982 through June 1983 — **10 months**. Peaked at 10.8% in November 1982. Missed the 12-month bar by 2 months. This is the post-WWII high-water mark.
- **2009 Great Recession**: unemployment touched 10.0% in October 2009 (Wikipedia/BLS show one source lists 10.1% peak), fell to 9.7% by January 2010, was below 10% by the next month. **One month** at the threshold.
- **2020 COVID**: spiked to 14.7% in April 2020, but back below 10% by August. **Two months** above 10%.
- Other above-10% single-month episodes: zero in the post-WWII series.

**Policy context (the response channel)**:
- **Trump EO 14365 (April 2026)**: limits AI regulation at federal and state level. California's Newsom signed a counter-EO March 2026 trying to preserve AI worker protections; expect a constitutional fight.
- **Sam Altman, May 2026** (Atlantic interview): walked away from UBI ("I no longer believe in universal basic income as much as I once did"), now favoring "shared ownership" / Public Wealth Fund model floated in OpenAI's "Industrial Policy for the Intelligent Era" paper.
- **Bannon faction in Trump orbit**: warning about AI-driven mass unemployment, pushing for protectionism. The Trump admin is internally split on whether AI displacement requires policy response.
- **State-level UI reform**: TCF report ("Our Unemployment System Needs Modernizing. Trump Is Doing the Opposite.") flags that current UI architecture would be slow to absorb a 5%+ displacement shock.
- **Historical compression mechanism**: In the 1981-82 and 2009 episodes, fiscal+monetary response (Reagan tax cuts, Obama ARRA, Fed cuts) and natural cyclical recovery brought unemployment below 10% within 10-12 months of peaking. CARES 2020 was even faster. For the 12-month sustained gate to fire, this compression mechanism has to fail — either because (a) the shock is too persistent (structural, not cyclical), (b) policy is too slow/inadequate, or (c) both.

**Where leading AI scenarios place 2026-2030 displacement**:
- **AI 2027 (Kokotajlo et al., Apr 2025)**: predicted significant job displacement starting late 2026. As of May 2026, AI 2027 Reality Tracker has 27 of 53 specific predictions confirmed/ahead/on-track (51%), with quantitative metrics at ~65% of predicted pace. Kokotajlo's own median timeline for AI automating >95% of remote jobs slipped from 2028 to 2029-30 in early 2026.
- **METR Time Horizon 1.1 (Jan 2026)**: post-2023 task-completion-time doubling rate 4.3 months (down from 7-month estimate). At 4.3-month doubling, an AI agent that could do a 30-minute task in early 2025 should be able to do a ~2-day task by late 2026 and a ~2-week task by mid-2027. That maps directly to the AI-agent-30pct-knowledge-work gate firing in the 2027-29 window.
- **Goldman Sachs (Mar 2026)**: AI reduced monthly US payroll growth by ~16K jobs over the prior year and added 0.1pp to unemployment. Base case: 6-7% gross workforce displacement over a decade, 0.6pp net unemployment impact. Frontloading would be much worse.
- **Acemoglu (NBER 2024, MIT Tech Review May 2026)**: AI lifts US TFP by 0.66% over a decade, 1.1% GDP. Implies near-zero net unemployment impact. He remains skeptical that agents replace messy multi-task human work.
- **Anthropic — Dario Amodei (May 2025 → May 2026)**: originally warned 10-20% unemployment within 5 years, half of entry-level white-collar jobs gone; explicitly walked back in May 2026, citing Jevons paradox ("if you automate 90% of the job, everyone does the 10%"). Implied his current view is closer to a 1-3pp unemployment bump, not 10%+.

**Net read of the current state**: the labor market in May 2026 is at 4.3% — not a crisis, but with multiple second-derivative warning signs (entry-level grad unemployment at 37-year high, hiring of young AI-exposed workers down 14%, labor force participation at multi-year low, tech layoffs running at 9x the 2025 AI-attributed rate). The cyclical channel (recession) and the structural channel (AI displacement) are both live but neither alone has the magnitude to clear the 12-consecutive-months >10% bar within the next 2-3 years. The gate fires within the decade *if and only if* both fire together, OR if AI displacement is several times faster than the median 2026 economist expects, OR if a Great-Depression-class shock occurs from an unrelated source (war, financial crisis, climate, infrastructure).

## Key uncertainties

1. **Speed of AI substitution vs. speed of reabsorption.** This is THE question. Acemoglu's 0.66% TFP-over-decade view implies essentially zero impact on the gate; Amodei's original 10-20% view implies the gate fires this decade. The 14% drop in young-AI-exposed-worker hiring since ChatGPT is a real signal that substitution is starting, but the official unemployment rate doesn't show it because (a) the substitution is currently faster than the *hiring* signal but not faster than the *separations* signal — i.e., firms aren't laying off existing workers, they're just not hiring new ones, and (b) labor force participation is absorbing some of the slack as discouraged workers exit. By 2028-2030, this should be more clearly resolved as the cohort effect compounds.

2. **Whether AI substitution is task-level or job-level (the Jevons question).** Amodei's May 2026 reframe — automate 90% of the task, but the human still does the 10% that requires judgment — implies AI productivity-multiplies labor rather than substituting for it. This is the optimistic Schumpeterian view. If correct, 30% of knowledge work being AI-handled (the upstream gate) does NOT translate to 30% of knowledge workers losing their jobs; instead, output per worker triples and the labor market reorganizes around higher-leverage roles. If wrong (the pessimistic / "winner-take-most" view), then a small number of AI-augmented humans capture the productivity gain and most knowledge workers are surplus.

3. **Policy response under a Republican-controlled federal government 2025-29.** Trump admin has rolled back AI regulation (EO 14365) and is hostile to expanded UI/UBI/safety net. If a sharp displacement shock arrives in 2026-28, the policy response will likely be slower and smaller than in 2009 (ARRA) or 2020 (CARES). State-level responses (Newsom CA EO March 2026) help but don't substitute. The probability of a 10%+ episode getting compressed back below threshold in <12 months drops materially under low-response policy regimes.

4. **Whether the recent METR doubling-rate acceleration is real and continues.** The Jan 2026 METR 1.1 update lowered the post-2023 doubling rate from ~7 months to ~4.3 months. If sustained, this implies AI agent capability multiplies ~10x every 14 months. If it slows (compute scaling hits diseconomies, algorithmic returns diminish, energy/cooling bottlenecks bite), the displacement timeline stretches by years. Most leading-edge model labs (OpenAI, Anthropic, Google DeepMind) are betting capability continues to scale; skeptics (Yann LeCun, Acemoglu) think we're near the top of an S-curve.

5. **The humanoid timeline.** Pure-cognition AI mostly only substitutes for the knowledge-work fraction (~60% of US jobs). Reaching sustained 10%+ unemployment from AI alone, without a recession, almost certainly requires humanoids substituting for physical labor too. The humanoid-retail-20k gate (P50 2028) and humanoid-self-replication-factory gate (P50 2034) are the rate-limiters here. If humanoids slip to mid-2030s, the structural-unemployment channel slips too.

6. **Labor force participation feedback.** If displaced workers exit the labor force (discouragement, early retirement, disability, education, caregiving), they don't count as "unemployed" by U-3 — but the lost output and the macro picture are still grim. The gate measures the official rate, so a "U-6 spikes to 15% but U-3 stays at 6% because participation collapses to 55%" scenario does NOT fire the gate. This is actually the more likely AI-disruption equilibrium per the post-2008 / opioid-crisis playbook — disengagement, not registered unemployment.

7. **A 'normal' recession deepening into something worse.** The 1981-82 episode missed the 12-month bar by 2 months. If a 2026-28 recession is even slightly deeper or slower-recovering than 1981-82, it clears the bar without needing any AI story. NY Fed term-spread model puts this at ~19% for the next ~18 months. If that recession compounds with the structural-displacement overlay, you get a genuinely Great-Depression-class episode — and the gate fires.

8. **Geopolitical / non-economic shocks.** A Taiwan invasion (and global semiconductor supply collapse), large-scale climate disaster, pandemic worse than COVID, or major US-China cold war escalation could trigger sustained double-digit unemployment via channels that have nothing to do with AI. These are tail risks but real; I'm folding them into the 2070 P90 anchor rather than modeling each separately.

## Sub-gate deep dives

### `us-recession-2026-2028` (P50 = 2027)

A standard cyclical NBER-defined recession in the 2026-2028 window. Current pricing: Polymarket/Kalshi at ~16% for 2026, ~45% for 2027; NY Fed term-spread model at ~19% for next 18 months. Sahm Rule indicator at 0.30, below the 0.50 trigger but elevated. Continuing claims have ticked up to 1.79M. The 2024 Sahm Rule trigger (0.54) didn't lead to recession — credit to the Fed's soft-landing engineering — but the labor market has weakened since.

If a recession arrives in 2026-27, the historical pattern suggests unemployment peaks 0-3 quarters AFTER the recession ends, in the 6-8% range for a "normal" recession or 9-10%+ for a deep one. **This sub-gate is necessary-but-insufficient** for the main gate. The 1981-82 episode produced 10 months above 10% from a baseline of 7.5%; getting to 12 months from a 2026-28 baseline of 4.3% requires either a deeper shock (banks fail, sovereign debt crisis, oil shock) or a slower recovery (policy gridlock, AI-substitution overlay preventing reabsorption).

### `us-unemployment-touches-10` (P50 = 2031)

A single monthly print of >=10%. Much easier than 12 consecutive months. Required as a stepping stone — you can't have 12 months above 10% without at least one. Metaculus prices this at ~1% by 2031, which I think is too low given Goldman/Amodei tail estimates and rising AI displacement signals. I'd put it at maybe 15-20% by 2031, with the median touch around 2031.

### `ai-displaces-5pct-workforce-net` (P50 = 2032)

Net displacement of 5%+ of US labor force = ~8M jobs net (gross AI displacement minus AI-enabled new roles minus reabsorption into other sectors). Current run-rate: CFO survey says ~502K AI-driven cuts in 2026, 0.4% of workforce, on track to be ~5% cumulative by 2030 if rate compounds at 50%/yr. Net (after reabsorption) is much lower in any historical model — closer to 2-3pp net unemployment impact even at full force per Goldman.

Hitting net 5% requires AI capability growth to outpace creative-destruction reabsorption durably. **Likeliest in the 2030-2035 window if METR doubling holds and humanoids commercialize.** Even at this level, you don't necessarily get 12-month-sustained-10% — you might get 7-8% sustained, which is bad but doesn't clear the gate.

### `labor-reabsorption-fails` (P50 = 2035)

The Schumpeterian / creative-destruction reabsorption mechanism that has reliably moved displaced US workers into new sectors since 1850 fails. Historically, sawyers became telephone operators became programmers became prompt engineers. The mechanism requires that new occupations being created (a) absorb a comparable number of workers, (b) within a few years, and (c) with reskilling that displaced workers can actually do. If AI is so general that any new task it creates is more efficiently done by AI itself, the reabsorption pipeline breaks. The ai-tutor-k8-parity-20mo gate firing helps (faster reskilling); the humanoid gates firing hurts (any new physical role is also robot-able).

This is a structural-economy gate, not a measurable single event. I'm placing it around 2035 as the median window where the labor-market academic literature would, in retrospect, identify the breakdown.

### `policy-fails-to-bridge-12mo` (P50 = 2036)

Even if structural unemployment spikes, the US has many policy levers to compress >10% episodes back below threshold within months — fiscal stimulus, expanded UI, public works, tariffs, monetary easing. Bush/Obama stimulus 2009-10 and CARES 2020 are recent precedents. For 12 consecutive months above 10%, the policy response must fail (political gridlock), arrive too late, or be insufficient.

Under current trajectory: Trump admin EO 14365 limits federal/state AI regulation; UI architecture has not been modernized; Sam Altman walked back UBI advocacy in May 2026; no major federal program addresses AI displacement. State responses (Newsom CA, EU AI Act, possibly UK/Israel) might lead but federal lag is the binding constraint.

A 2036 P50 reflects: the gate fires only in worlds where policy is meaningfully slower than the 2009 or 2020 responses, which is plausible under continued political polarization but is not the modal expectation.

## Cross-gate interactions

(See `cross_gate` block above for the structured list.)

**Strongest enabler**: `ai-agent-30pct-knowledge-work` (P50 2029). If AI agents handle 30%+ of knowledge-work tasks by 2029, the substitution-channel inputs to mass unemployment are firmly in place. Without this gate firing, the structural-unemployment thesis loses its primary mechanism — and the main gate reverts to "ordinary recession plus policy failure" scenarios with much lower probability.

**Strongest substitute / inverse**: `global-economy-explosive-growth` (P50 2049, canonical 2x tier). If global GDP doubles, the labor demand for non-substitutable work explodes — making sustained 10%+ unemployment less likely, not more. The two gates can both fire, but the explosive-growth path makes the main gate less likely conditional on AI-capability gates firing. The worst case for unemployment is "AI productivity surge without GDP boom landing" — a productivity-without-growth equilibrium where capital captures all the upside.

**Stepping-stone gates**: `humanoid-retail-20k` (P50 2028) → extends substitution from white-collar to physical service work. `humanoid-self-replication-factory` (P50 2034) → makes embodied labor effectively unbounded in supply at marginal cost. Without both, AI-driven unemployment plateaus around the knowledge-work fraction.

**Reskilling counter-force**: `ai-tutor-k8-parity-20mo` (P50 2031) → strengthens reabsorption. If displaced workers can be retrained rapidly via AI tutors, the policy-bridge sub-gate becomes much less likely to fire.

## Sources

1. [BLS Employment Situation Summary, 2026 M04 Results](https://www.bls.gov/news.release/empsit.nr0.htm)
2. [Indeed Hiring Lab — US Labor Market Snapshot April 2026](https://www.hiringlab.org/2026/05/14/us-labor-market-snapshot-april-2026/)
3. [BLS Civilian Unemployment Rate chart](https://www.bls.gov/charts/employment-situation/civilian-unemployment-rate.htm)
4. [FRED UNRATE series](https://fred.stlouisfed.org/series/UNRATE)
5. [BLS Monthly Labor Review: "Unemployment continued to rise in 1982"](https://www.bls.gov/opub/mlr/1983/02/art1full.pdf)
6. [Washington Post archive: "Unemployment Rate Rises to 10.8%" (Dec 1982)](https://www.washingtonpost.com/archive/politics/1982/12/04/unemployment-rate-rises-to-108/c8668245-d58f-4646-9de9-e4d87be74220/)
7. [Federal Reserve History: Recession of 1981-82](https://www.federalreservehistory.org/essays/recession-of-1981-82)
8. [BLS Spotlight on Statistics: The Recession of 2007-2009](https://www.bls.gov/spotlight/2012/recession/pdf/recession_bls_spotlight.pdf)
9. [Berkeley IRLE: High Unemployment after the Great Recession (June 2010)](https://irle.berkeley.edu/wp-content/uploads/2010/06/High-Unemployment-After-the-Great-Recession.pdf)
10. [Anthropic Economic Index report: Learning Curves (March 2026)](https://www.anthropic.com/research/economic-index-march-2026-report)
11. [Anthropic Economic Index report: Economic Primitives (January 2026)](https://www.anthropic.com/research/anthropic-economic-index-january-2026-report)
12. [Nexford: What Anthropic's 2026 AI Labor Market Report Means for Your Career](https://www.nexford.edu/insights/what-anthropics-2026-ai-labor-market-report-means-for-your-career)
13. [Goldman Sachs: How Will AI Affect the US Labor Market?](https://www.goldmansachs.com/insights/articles/how-will-ai-affect-the-us-labor-market)
14. [Goldman Sachs: Jobs AI is likely to boost — and disrupt](https://www.goldmansachs.com/insights/articles/the-jobs-ai-is-likely-to-boost-and-those-it-may-disrupt)
15. [AI2Work: Goldman Sachs says AI job losses could force Fed rate cuts](https://ai2.work/blog/goldman-sachs-says-ai-job-losses-could-force-the-fed-to-cut-rates)
16. [Acemoglu: The Simple Macroeconomics of AI (NBER w32487)](https://www.nber.org/system/files/working_papers/w32487/w32487.pdf)
17. [MIT Tech Review: Three things in AI to watch (Acemoglu, May 2026)](https://www.technologyreview.com/2026/05/11/1137090/three-things-in-ai-to-watch-according-to-a-nobel-winning-economist/)
18. [Axios: Top AI CEO foresees white-collar bloodbath (Amodei, May 2025)](https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic)
19. [Fortune: Amodei walking back AI white-collar bloodbath (May 2026)](https://fortune.com/2026/05/05/dario-amodei-jevons-paradox-will-ai-wipe-out-white-collar-jobs/)
20. [Fortune: Altman/Amodei walking back AI job apocalypse ahead of IPOs (May 2026)](https://fortune.com/2026/05/26/sam-altman-dario-amodei-walking-back-ai-jobs-apocalypse-prophecies-ipo/)
21. [AI 2027 (Kokotajlo et al.)](https://ai-2027.com/)
22. [AI 2027 Reality Tracker](https://ai2027-tracker.com/)
23. [LessWrong: AI 2027 Tracker — one year of predictions vs reality](https://www.lesswrong.com/posts/oSWae4bE4mqWy5a6Q/ai-2027-tracker-one-year-of-predictions-vs-reality)
24. [Futuresearch: AI 2027 Six Months Later](https://futuresearch.ai/ai-2027-6-months-later/)
25. [METR Time Horizon 1.1 (January 2026)](https://metr.org/blog/2026-1-29-time-horizon-1-1/)
26. [METR: Task-Completion Time Horizons of Frontier AI Models](https://metr.org/time-horizons/)
27. [AI Digest: A New Moore's Law for AI Agents](https://theaidigest.org/time-horizons)
28. [Kalshi (Trader's Union summary): higher US unemployment risk priced as AI concerns grow](https://tradersunion.com/news/financial-news/show/2138253-ai-unemployment-risk-kalshi-market/)
29. [Metaculus: US unemployment rate 10% before 2031](https://www.metaculus.com/questions/18664/us-unemployment-rate-10-before-2031/)
30. [Metaculus Labor Automation Forecasting Hub](https://www.metaculus.com/labor-hub/)
31. [Manifold: Will AI cause US unemployment > 10% before 2030?](https://manifold.markets/ahalekelly/will-ai-cause-the-us-unemployment-r)
32. [Manifold: Will AI cause mass unemployment by 2030?](https://manifold.markets/AitchKay/will-ai-cause-mass-unemployment-by)
33. [CNBC: 20K Meta/Microsoft cuts raise concern AI-driven labor crisis is here (Apr 2026)](https://www.cnbc.com/2026/04/24/20k-job-cuts-at-meta-microsoft-raise-concern-of-ai-labor-crisis-.html)
34. [Tom's Hardware: Tech industry lays off ~80K in Q1 2026, ~50% AI-attributed](https://www.tomshardware.com/tech-industry/tech-industry-lays-off-nearly-80-000-employees-in-the-first-quarter-of-2026-almost-50-percent-of-affected-positions-cut-due-to-ai)
35. [TechTimes: Tech layoffs reach 142,000 in 2026 to fund $700B AI infrastructure](https://www.techtimes.com/articles/317392/20260529/tech-layoffs-reach-142000-2026-profitable-companies-cut-jobs-fund-700b-ai-infrastructure.htm)
36. [24/7 Wall St: $725B going to four AI companies (May 2026)](https://247wallst.com/investing/2026/05/07/tens-of-thousands-of-tech-workers-are-being-laid-off-in-2026-the-725-billion-that-replaced-them-is-going-to-four-companies/)
37. [Washington Post: Layoffs at Amazon, Meta, Microsoft aren't all about AI (May 2026)](https://www.washingtonpost.com/technology/2026/05/01/ai-jobs-tech-layoffs-austerity/)
38. [Invezz: Is Big Tech's $725B AI splurge being funded by mass layoffs?](https://invezz.com/news/2026/05/04/is-big-techs-725b-ai-splurge-being-funded-by-mass-layoffs/)
39. [TechJackSolutions / NBER summary: CFO survey projects 502K AI-driven cuts in 2026](https://techjacksolutions.com/ai-brief/nber-survey-of-750-cfos-projects-ai-driven-job-cuts-will-rea/)
40. [Fortune: CFOs admit AI layoffs will be 9x higher this year (March 2026)](https://fortune.com/2026/03/24/cfo-survey-ai-job-cuts-productivity-paradox-2026/)
41. [Tom's Hardware: 99% of CEOs expect AI-driven layoffs (Mercer 2026 Global Talent Trends)](https://www.tomshardware.com/tech-industry/artificial-intelligence/survey-reveals-that-99-percent-of-ceos-now-expect-ai-driven-layoffs)
42. [Fortune: AI isn't paying off the way companies think (Gartner study)](https://fortune.com/2026/05/11/ai-automation-layoffs-gartner-study-roi/)
43. [Fortune: Entry-level AI elimination + 1962 Nobelist Arrow (May 2026)](https://fortune.com/2026/05/21/kenneth-arrow-learning-by-doing-entry-level-work-automation/)
44. [Washington Monthly: How AI Broke the Entry-Level Job (May 2026)](https://washingtonmonthly.com/2026/05/29/ai-entry-level-jobs-college-graduates/)
45. [FastApply blog: Entry-level jobs disappearing 2026, 9.7% grad unemployment](https://blog.fastapply.co/entry-level-jobs-disappearing-2026)
46. [Bloomberg: Why more college graduates are stuck in jobs not requiring degrees (April 2026)](https://www.bloomberg.com/news/articles/2026-04-13/ai-and-a-frozen-job-market-leave-college-grads-underemployed)
47. [Bloomberg: US jobless claims edge up to 215K (May 28, 2026)](https://www.bloomberg.com/news/articles/2026-05-28/us-jobless-claims-increase-slightly-to-highest-in-over-a-month)
48. [San Francisco Fed: Recent Slowdown in Labor Supply and Demand (Jan 2026)](https://www.frbsf.org/research-and-insights/publications/economic-letter/2026/01/recent-slowdown-in-labor-supply-and-demand/)
49. [Indeed Hiring Lab: Labor Force Participation projected to fall through 2034](https://www.hiringlab.org/2026/04/07/why-labor-force-participation-is-projected-to-fall-through-2034/)
50. [FRED Sahm Rule SAHMREALTIME](https://fred.stlouisfed.org/series/SAHMREALTIME)
51. [Motley Fool: Sahm Rule flashing again (Feb 2026)](https://www.fool.com/investing/2026/02/21/this-signal-coincided-with-recession-past-65-years/)
52. [CurrentMarketValuation: Sahm Rule recession indicator](https://www.currentmarketvaluation.com/models/sahm-rule.php)
53. [Deloitte US Economic Forecast Q1 2026](https://www.deloitte.com/us/en/insights/topics/economy/us-economic-forecast/united-states-outlook-analysis.html)
54. [Money: Will We Have a Recession in 2026? Most Economists Say No](https://money.com/recession-predictions-2026/)
55. [Polymarket: US recession by end of 2026](https://polymarket.com/event/us-recession-by-end-of-2026)
56. [Stanford SIEPR: US economy in 2026 — what to watch for](https://siepr.stanford.edu/publications/policy-brief/us-economy-2026-what-watch)
57. [Federal Reserve Board: Labor force growth, breakeven employment, potential GDP growth (April 2026)](https://www.federalreserve.gov/econres/notes/feds-notes/labor-force-growth-breakeven-employment-and-potential-gdp-growth-20260402.html)
58. [Kansas City Fed: New monthly estimates of natural rate of unemployment](https://www.kansascityfed.org/research/economic-bulletin/introducing-new-monthly-estimates-of-the-natural-rate-of-interest-and-natural-unemployment-rate/)
59. [Fed Vice Chair Jefferson: speech on economic outlook and labor market (April 2026)](https://www.federalreserve.gov/newsevents/speech/jefferson20260407a.htm)
60. [Yahoo Finance / Atlantic: Sam Altman puts $14M into UBI study, now says UBI not what we need](https://finance.yahoo.com/economy/policy/articles/sam-altman-put-14m-studying-160110092.html)
61. [Asia Business Daily: Altman says sharing AI wealth needed over UBI](https://www.asiae.co.kr/en/article/world-general/2026050615071397815)
62. [Startup Fortune: Sam Altman has changed his mind about UBI](https://startupfortune.com/sam-altman-has-changed-his-mind-about-universal-basic-income-and-the-reasoning-matters-more-than-the-headline/)
63. [MIT Tech Review: America's coming war over AI regulation](https://www.technologyreview.com/2026/01/23/1131559/americas-coming-war-over-ai-regulation/)
64. [Law and the Workplace: Trump AI EO 14365 (April 2026)](https://www.lawandtheworkplace.com/2026/04/what-president-trumps-ai-executive-order-14365-means-for-employers/)
65. [California Governor's Office: Newsom signs EO to strengthen AI worker protections (March 2026)](https://www.gov.ca.gov/2026/03/30/as-trump-rolls-back-protections-governor-newsom-signs-first-of-its-kind-executive-order-to-strengthen-ai-protections-and-responsible-use/)
66. [TCF: Our Unemployment System Needs Modernizing. Trump Is Doing the Opposite](https://tcf.org/content/report/our-unemployment-system-needs-modernizing-trump-is-doing-the-opposite/)
67. [Jerusalem Post: AI ending 'job immunity' for Israel's young tech workers](https://www.jpost.com/business-and-innovation/all-news/article-894745)
68. [Calcalist: Full list of Israeli high-tech layoffs in 2026](https://www.calcalistech.com/ctechnews/article/fgffufeb3)
69. [Times of Israel: Major Israeli tech firms commence sweeping layoffs as AI roils industry](https://www.timesofisrael.com/major-israeli-tech-firms-commence-sweeping-layoffs-as-ai-revolution-roils-industry/)
70. [Times of Israel: Wix said to be cutting ~20% of global workforce](https://www.timesofisrael.com/liveblog_entry/israeli-tech-giant-wix-said-to-be-cutting-some-20-of-global-workforce-in-coming-months/)
71. [BLS Monthly Labor Review: Great Recession, Great Recovery (CPS trends)](https://www.bls.gov/opub/mlr/2018/article/great-recession-great-recovery.htm)
72. [Cleveland Fed: Recessions and the Trend in the US Unemployment Rate](https://www.clevelandfed.org/publications/economic-commentary/2021/ec-202101-recessions-and-the-trend-in-the-us-unemployment-rate)
73. [Richmond Fed: Business cycles — is current environment different? (2026)](https://www.richmondfed.org/publications/research/economic_brief/2026/eb_26-07)
74. [CBO Budget and Economic Outlook 2026-2036](https://www.cbo.gov/publication/61882)
75. [CBO: Additional Information About the Economic Outlook 2025-2035](https://www.cbo.gov/publication/61189)
76. [Macrotrends: US Unemployment Rate 1948-2026](https://www.macrotrends.net/1316/us-national-unemployment-rate)