AI and Unemployment [Mar 2026]
Will AI cause mass unemployment? The range of expert predictions spans from a temporary +0.5 percentage point rise (Goldman Sachs) to 10–20% unemployment rivalling the Great Depression (Anthropic’s CEO). We compiled 80 data points from 59+ sources, scored 3649 individual roles against real AI capabilities, and mapped the results to BLS employment data covering 168.7M US workers.
516 roles sit in the RED zone — representing 44.3M US workers in occupations where AI can already perform a significant portion of core tasks. Below, we walk through what’s actually happening, who’s getting hit first, the historical parallels, the freelancer canary in the coal mine, and why the safety net isn’t ready.
🎯 The Short Answer
AI is not causing mass unemployment — yet. The US unemployment rate sits at 4.28%, well within normal range. But beneath that headline number, displacement is already measurable: 55,000 US job losses attributed to AI in 2025, a 0.47 correlation between AI exposure and rising unemployment across occupations, and 37% of companies planning to replace workers with AI by year-end. The question is not whether AI displaces workers — the data already shows it does. The question is how fast and how far.
The headline unemployment rate masks sectoral disruption. 516 roles in our database — covering 44.3M US workers — already sit in the RED zone where AI can perform a significant portion of core tasks. The gap between this measured risk and the headline 4.28% rate is where the story lives.
📊 What’s Actually Happening Right Now
The gap between measured AI job losses (55,000 in 2025) and planned workforce reductions (hundreds of thousands across major corporations) is where the risk lives. Companies are making hiring decisions based on AI’s potential, not its current performance. 29% are already hiring fewer people in anticipation of AI capabilities, while only 12% of US workers actually use generative AI daily.
The St. Louis Federal Reserve found a 0.47 correlation between AI exposure and unemployment increases across occupations from 2022 to 2025. Computer and mathematical roles — with ~80% AI exposure — saw the largest unemployment gains. Yale’s Budget Lab, however, found a slight employment increase in high-AI-exposure roles through late 2025. The data is pulling in two directions simultaneously.
| Finding | Value | Source |
|---|---|---|
| US headline unemployment rate (early 2026) | 4.28% | BLS / Citadel Securities |
| US job losses attributed to AI in 2025 | 55,000 | Challenger, Gray & Christmas |
| AI’s share of all US job cuts (2025) | 4.5% | Challenger, Gray & Christmas |
| Correlation between AI exposure and rising unemployment across occupations (US) | 0.47 | Federal Reserve Bank of St. Louis (Aug 2025) |
| Employment change in high-AI-exposure roles, mid-2024 to late 2025 (US) | +1.7% | Yale Budget Lab (Jan 2026) |
| Companies hiring fewer people in anticipation of AI (US) | 29% | HBR (Jan 2026) |
| Companies planning to replace staff with AI by end of 2026 (US) | 37% | Resume.org (1,000 US leaders) |
| Organizations that have already made large AI-driven headcount cuts (US) | 2% | HBR (Jan 2026) |
| Employers expecting AI headcount reductions this year (US) | 1 in 6 | Industry surveys |
| US workers who use generative AI daily at work | 12% | St. Louis Fed |
The gap between measured job losses (55,000) and planned cuts (hundreds of thousands) is where the risk lives. Companies are making hiring decisions based on AI’s potential capabilities, not its current deployed performance.
AI-attributed job losses in 2025 totalled roughly 55,000 (4.5% of all US job cuts). But 37% of companies plan to replace workers with AI by end of 2026. The planned number is orders of magnitude larger than the measured one. Either the plans will be scaled back, or the measured losses are about to accelerate.
🔮 Will AI Cause Mass Unemployment?
Economists and AI researchers disagree sharply on trajectory. Goldman Sachs models a temporary +0.5 percentage point unemployment rise that resolves within two years. Anthropic’s CEO warns of 10–20% unemployment — rivalling the Great Depression. The disagreement is not about whether AI displaces workers. It’s about speed of adoption and speed of adaptation.
The Optimists
Goldman Sachs (Aug 2025): Temporary US unemployment rise of +0.5 percentage points, resolving within 2 years. A 15% labour productivity boost offsets displacement.
Displacement range: 6–7% of the US workforce displaced (3–14% range), with most workers transitioning to adjacent roles.
PwC AI Jobs Barometer: AI-exposed industries seeing 2× faster wage growth — augmentation over replacement.
Federal Reserve: Projects 4.4% unemployment by end of 2026 — no AI-driven spike.
Model: gradual adoption. Historical pattern of technology creating more jobs than it destroys.
The Pessimists
Dario Amodei (Anthropic CEO): 10–20% US unemployment in 1–5 years. “It could feel like a depression.”
Geoffrey Hinton (Turing Award): “Massive unemployment is very likely.”
JP Morgan: 3–6% US displacement in 1–3 years, 10–15% over a decade.
McKinsey: 57% of US work hours are automatable with current AI technology.
Model: rapid capability jumps. AI advancing faster than the workforce can adapt.
The Middle Ground
🌍 WEF (2025): +170M jobs, −92M displaced = net +78M globally by 2030
🇺🇸 McKinsey: 12M occupational transitions needed in the US by 2030
🌍 IMF: 40% of global jobs exposed; 60% in advanced economies
🇺🇸 MIT (Nov 2025): AI can already complete tasks of ~12% of the US workforce
These are not minor disagreements. Goldman Sachs is modelling gradual adoption — the way every previous technology played out. Amodei and Hinton are modelling rapid capability jumps — something we have not seen before. The answer depends on which model proves correct.
| Finding | Value | Source |
|---|---|---|
| Goldman Sachs: modelled temporary rise in US unemployment | +0.5pp | Goldman Sachs (Aug 2025) |
| Goldman Sachs: estimated share of US workforce displaced | 6–7% (range 3–14%) | Goldman Sachs (Aug 2025) |
| Anthropic CEO: possible US unemployment from AI | 10–20% | Dario Amodei (May 2025) |
| JP Morgan: US displacement expected within 1–3 years | 3–6% | JP Morgan Private Bank |
| IMF: share of global jobs exposed to AI | 40% | International Monetary Fund (2024) |
| IMF: jobs exposed in advanced economies (Global) | 60% | International Monetary Fund (2024) |
| WEF: jobs displaced worldwide by 2030 | 92 million | World Economic Forum (2025) |
| WEF: new jobs created worldwide by 2030 | 170 million | World Economic Forum (2025) |
| McKinsey: US workers needing occupational transitions by 2030 | 12 million | McKinsey Global Institute |
| MIT: share of the US workforce whose tasks AI can already perform | ~12% | MIT (Nov 2025) |
| OECD: jobs in occupations where 50%+ of tasks are automatable (Global) | 27% | OECD Employment Outlook 2023 |
| PwC: share of jobs automatable by the mid-2030s (Global) | Up to 30% | PwC |
| Goldman Sachs: full-time jobs exposed to generative AI globally | 300 million | Goldman Sachs |
| Geoffrey Hinton: likelihood of massive unemployment (Global) | "Very likely" | Geoffrey Hinton (Nobel Prize, Bloomberg TV Nov 2025) |
| McKinsey: share of US work hours automatable with AI | 57% | McKinsey Global Institute (2025) |
| Federal Reserve: projected unemployment rate end of 2026 (US) | 4.4% | Federal Reserve |
The range between Goldman’s +0.5pp and Amodei’s 10–20% represents a 20x disagreement between serious institutions. This is not normal for economic forecasting. It reflects genuine uncertainty about the speed of AI capability growth and adoption.
🏢 Corporate AI Layoffs — Who’s Cutting
The largest AI-attributed workforce reductions are concentrated in technology and finance. Amazon cut 30,000 corporate roles across two rounds. Wall Street banks plan ~200,000 cuts over 3–5 years. Salesforce removed 4,000 positions citing AI productivity. These are not hypothetical — they are board-level decisions already underway.
Amazon — 14,000 corporate roles (Oct 2025), then 16,000 more (Jan 2026). Largest single-company AI-driven reductions to date.
Salesforce — 4,000 roles (Sep 2025). CEO stated AI handles roughly half the company’s workload.
Wall Street banks — Planning ~200,000 cuts across major institutions over 3–5 years, citing AI-driven automation of back-office functions (Bloomberg).
BT Group — Up to 55,000 job cuts planned, citing AI and automation as the primary driver.
A critical nuance: 60% of companies that cited AI as a reason for layoffs were actually using AI as a scapegoat for unrelated cost-cutting. The real AI-driven reductions are concentrated in a smaller number of companies that have genuinely deployed the technology.
| Finding | Value | Source |
|---|---|---|
| Companies that have already replaced workers with AI (US) | 30% | Resume.org (1,000 US leaders) |
| Companies that used AI as a scapegoat for unrelated layoffs (US) | 59% | Resume.org (1,000 hiring managers) |
| Employers planning workforce reduction due to AI by 2030 (Global) | 40% | World Economic Forum |
| BT Group: planned job cuts citing AI and automation (UK) | 55,000 total / 10,000 AI-replaced | BT Group (2023) |
| US job cuts announced in January 2026 | 108,435 | Challenger, Gray & Christmas |
Not every “AI layoff” is an AI layoff. Research shows 60% of companies that cited AI were using it as justification for cuts driven by other factors. When assessing AI’s true employment impact, separate the signal from the noise.
🧠 The Anticipation Gap — Potential vs Performance
Harvard Business Review’s January 2026 analysis identified a critical pattern: companies are laying off based on AI’s potential, not its actual performance. 77% of recent AI-attributed layoffs were anticipatory — the technology hadn’t yet replaced the workers. This matters because premature cuts may be reversed if AI underdelivers, or accelerated if it overdelivers.
This matters because it means some of the displacement we are seeing is not driven by AI actually doing the work — it is driven by executives betting that AI will do the work. If that bet is wrong, some of these job cuts are premature cost-cutting dressed up as AI transformation.
The adoption data tells a more nuanced story: 72% of organizations now use AI in at least one business function, but only 1% report mature deployment. 12% of US workers use GenAI daily at work. The gap between organizational experimentation and actual production deployment is wide.
| Finding | Value | Source |
|---|---|---|
| AI-attributed layoffs that were anticipatory, not performance-based (US) | 77% | HBR (Jan 2026) |
| Companies reducing hiring in anticipation of AI (US) | 29% | HBR (Jan 2026) |
| Organizations with large AI-driven workforce reductions (US) | 2% | HBR (Jan 2026) |
| US workers actually using GenAI daily | 12% | St. Louis Fed |
| Companies reporting mature AI deployment (Global) | 1% | McKinsey State of AI (2024) |
| Organizations using AI in at least one business function (Global) | 78% | McKinsey (2025) |
Companies are cutting headcount in anticipation of AI capabilities while simultaneously reporting that only 1% have achieved mature AI deployment. This creates a paradox: the workforce reductions are running ahead of the technology adoption that would justify them.
⚠️ Who Gets Hit First
Displacement is not evenly distributed. Young workers, women, and those in clerical roles bear disproportionate risk. Stanford found a 13% employment decline for 22–25 year-olds in AI-exposed occupations since November 2022. The IMF reports 9.6% of female employment in high-income countries faces high vulnerability — nearly 3x the 3.2% rate for men.
Young Workers
Stanford: 13% employment decline for 22–25 year-olds in AI-exposed occupations since Nov 2022. Big Tech cut new-grad hiring by 25%. Youth unemployment (ages 20–24) hit 9.5% in Sep 2025 — more than double the headline rate.
Women
IMF: 9.6% of female employment in high-income countries faces high AI vulnerability — 3x the 3.2% rate for men. Women are overrepresented in clerical and administrative roles where AI exposure is highest.
Low-Adaptive Workers
Brookings: 6.1 million workers with high AI exposure and low adaptive capacity — limited savings, low skill transferability, narrow reemployment prospects. 86% concentrated in clerical roles in smaller metro areas.
Entry-Level Roles
Anthropic’s CEO predicts 50% displacement of entry-level white-collar tasks within 1–5 years. New college grad unemployment is already ~10%. The entry-level pipeline — where workers build skills — is narrowing.
For the full breakdown on entry-level impact, see Is AI Replacing Entry Level Jobs?
| Finding | Value | Source |
|---|---|---|
| Employment decline for 22–25 year-olds in AI-exposed US occupations since Nov 2022 | -16% | Stanford DEL (Brynjolfsson et al., 2025) |
| Big Tech new-graduate hiring reduction, 2023–2024 (US) | -25% | Goldman Sachs (2025) |
| Female employment vulnerability to AI automation, vs 3.2% for men (Global) | 9.6% | IMF (2024) |
| Workers with high AI exposure and low capacity to adapt (US) | 6.1 million | Brookings Institution (2026) |
| Global employees who fear losing their job to AI (2024 → 2026) | 28% → 40% | Mercer (12,000 respondents) |
| US workers more worried than hopeful about AI at work | 52% | Pew Research (Oct 2024) |
| Workers who say AI job elimination is “not at all likely” (US) | 50% (down from 60% in 2023) | Gallup (2025) |
| New college graduate unemployment rate, early 2026 (US) | ~10% | Goldman Sachs / Industry data |
| US youth unemployment rate ages 20–24 (Sep 2025) | 9.5% | BLS / Fortune |
| Drop in young workers’ ability to find employment in AI-exposed roles (US) | -14% | Anthropic Research (2025) |
| Anthropic CEO: timeline for 50% displacement of entry-level white-collar tasks (US) | 50% within 1–5 years | Dario Amodei (May 2025) |
Young workers face a compounding problem: fewer entry-level roles available, higher competition for those that remain, and a shrinking window to build the human skills (judgement, relationships, domain expertise) that protect against AI displacement. The traditional career ladder is losing its first rung.
📈 Our Data — 3649 Roles Scored
Our AIJRI scoring framework assessed each role task-by-task against current AI capabilities — not surveys, not sector averages, not assumptions. The result: a granular map of which specific roles face displacement and how many workers hold them.
The 20 most vulnerable roles in our database, ranked by JobZone Score (lowest first):
| # | Role | Score |
|---|---|---|
| 1 | File Clerks (Mid-Level) | 1.5 /100 |
| 2 | Micro-Task Worker (Online) (Mid-Level) | 1.7 /100 |
| 3 | Data Entry Keyer (Mid-Level) | 2.3 /100 |
| 4 | Word Processor and Typist (Mid-Level) | 2.6 /100 |
| 5 | Vulnerability Tester / Scanner Operator (Entry-Level) | 2.7 /100 |
| 6 | Telephone Operator (Mid-Level) | 3.0 /100 |
| 7 | Virtual Assistant (Entry-to-Mid Level) | 3.2 /100 |
| 8 | Live Chat Support Agent (Entry-to-Mid Level) | 3.4 /100 |
| 9 | Telemarketer (Mid-Level) | 3.4 /100 |
| 10 | Medical Transcriptionist (Mid-Level) | 3.6 /100 |
| 11 | Toll Collector (Mid-Level) | 3.6 /100 |
| 12 | Machine Feeders and Offbearers (Mid-Level) | 3.6 /100 |
| 13 | Procurement Clerks (Mid-Level) | 3.6 /100 |
| 14 | Correspondence Clerk (Mid-Level) | 3.6 /100 |
| 15 | Desktop Publisher (Mid-Level) | 3.7 /100 |
| 16 | Office Machine Operator, Except Computer (Mid-Level) | 3.9 /100 |
| 17 | OnlyFans Chatter / Ghostwriter (Entry-to-Mid Level) | 4.0 /100 |
| 18 | Meter Reader (Mid-Level) | 4.1 /100 |
| 19 | Medical Scribe (Mid-Level) | 4.3 /100 |
| 20 | Insurance Claims and Policy Processing Clerk (Entry-to-Mid) | 4.4 /100 |
RED zone employment: 44.3M US workers hold roles in the RED zone. These are not predictions — they are current assessments based on what AI can do today.
The roles at the bottom of this list operate in digital-first environments where AI tools already handle core workflows. See the full list: Jobs Most at Risk From AI.
Our scoring is task-level, not opinion-level. Each role was assessed against 40+ specific AI capability dimensions. 516 roles in the RED zone means 44.3M workers in occupations where AI can already perform a significant portion of the work. This is not a forecast — it is a measurement of current capability overlap.
🏭 Sectors Most Exposed
AI displacement risk varies dramatically by sector. Domains with predominantly digital, repetitive, or data-processing workflows score lowest on AI resistance. Domains requiring physical presence, licensing, or high-stakes human judgement score highest.
The pattern is clear: domains with predominantly digital, repetitive, or data-processing workflows cluster at the bottom. Domains requiring physical presence, professional licensing, or high-stakes human judgement cluster at the top. This is not about “low-skill” vs “high-skill” — some highly skilled digital roles (software engineering, data science) score lower than skilled trades.
Most Exposed Sectors — Role Breakdown
For a broader view of which industries face the most AI exposure, see AI and Job Loss Statistics.
The dividing line is not education level or salary — it is whether the role’s core tasks happen in physical space or digital space. A plumber with no degree scores higher than a data analyst with a master’s. AI is strongest where work lives entirely in software.
🌍 The Global Picture
AI displacement is not a US-only phenomenon. The IMF estimates 40% of global jobs are exposed to AI — rising to 60% in advanced economies. Country-level unemployment rates tell different stories, but the structural exposure is global.
United States
Unemployment: 4.28%
AI job losses (2025): 55,000
RED zone workers: 44.3M
McKinsey: 12M transitions by 2030
United Kingdom
Unemployment: 4.4%
Workers fearing AI job loss: 25%+
AI identified as hitting UK harder than US
Businesses using AI (2023): ~15%
European Union
Unemployment: 5.9%
Youth unemployment: 14.3%
Germany: 6.0%
OECD average: 4.9%
Canada
Unemployment: 6.7%
Australia
Unemployment: 4.1%
India
Unemployment: 3.2%
Youth unemployment: 10.2%
Global unemployment stands at 5.0% (ILO). The ILO projects total global employment at 3.5 billion. Youth unemployment globally runs at 13% — well above the headline rate in every region. AI displacement risk adds a new layer on top of existing structural unemployment.
| Finding | Value | Source |
|---|---|---|
| Global unemployment rate (Global, ILO) | 5.0% | ILO World Employment & Social Outlook 2025 |
| Total global employment (Global, ILO) | 3.44B | ILO World Employment & Social Outlook 2025 |
| Global youth unemployment rate, 15–24 (Global) | 13.0% | ILO World Employment & Social Outlook 2025 |
| UK unemployment rate | 4.4% | ONS Labour Market Overview |
| German unemployment rate | 6.0% | Destatis / Bundesagentur für Arbeit |
| EU unemployment rate | 5.9% | Eurostat Unemployment Statistics |
| EU youth unemployment rate (under 25) | 14.3% | Eurostat Youth Employment Statistics |
| Canadian unemployment rate | 6.7% | Statistics Canada Labour Force Survey |
| Australian unemployment rate | 4.1% | ABS Labour Force |
| India unemployment rate, PLFS (India) | 3.2% | PLFS Annual Report 2024-25 |
| OECD average unemployment rate, July 2025 (Global) | 4.9% | OECD |
| UK workers fearing job loss to AI within 5 years (UK) | 25%+ | Randstad Survey (Jan 2026) |
| UK businesses that had adopted AI by 2023 | ~15% | ONS |
| Share of work tasks employers predict will be automated by 2030 (Global) | 34% | Tenet / Employer Surveys |
UK and Global figures are proportional estimates based on our assessed US zone distribution. Actual exposure may differ due to industry mix and regulatory differences.
A key finding from PwC: wages in AI-exposed industries are growing twice as fast as in non-exposed industries. That suggests AI is augmenting workers in many roles, not just replacing them. The net effect on employment is not a simple subtraction.
The IMF’s finding that 60% of jobs in advanced economies are exposed to AI — versus 40% globally — reflects a structural reality: advanced economies have more office-based, digital, and knowledge-worker roles. Developing economies with larger agricultural and manual labour sectors face lower AI exposure but also lower AI productivity gains.
📜 Has This Happened Before?
Every major wave of automation has triggered the same fear: mass unemployment. The historical record is clear — and complicated. Previous waves (ATMs, manufacturing robots, personal computers) consistently created more jobs than they destroyed. But AI differs in three ways: it’s cognitive, the speed is faster, and the scope is broader.
ATMs and Bank Tellers
When ATMs arrived in the 1970s, the prediction was that tellers would disappear. US bank tellers grew from 300,000 (1980) to nearly 600,000 (2010). ATMs reduced the cost per branch — banks opened more branches — and hired more tellers for customer service and sales. The job title stayed. The work changed.
Source: James Bessen, Boston University, 2015
140 Years of Evidence
Deloitte analysed 140 years of census data from England and Wales. Technology has been a “great job-creating machine.” Agriculture and manufacturing employment collapsed, but was more than offset by growth in care, creative, tech, and business services. 60% of US workers today are in occupations that did not exist in 1940.
Source: Deloitte, 2015; Goldman Sachs Research
Why AI Might Be Different
Three things separate AI from every previous wave. First, it is cognitive, not manual — AI automates thinking, not just doing. Second, the speed — adoption is outpacing reskilling capacity. Third, the scope — AI affects creative, analytical, and expert knowledge tasks previously considered safe. MIT economist David Autor found automation is “hollowing out” the middle class.
Source: Autor, MIT, 2015; Goldman Sachs, 2025
For Context: US Unemployment Peaks
| Crisis | Peak Rate | Year |
|---|---|---|
| Great Depression | ~25% | 1933 |
| COVID-19 pandemic | 14.8% | 2020 |
| Global Financial Crisis | 10% | 2009 |
| Current (early 2026) | 4.28% | 2026 |
| Amodei’s worst case | 10–20% | 1–5 years |
Source: NBER, BLS, Federal Reserve Bank of St. Louis
In 2024, AI directly created 119,900 US jobs (8,900 AI development + 110,000 data centre construction) against 12,700 losses — a net gain of 107,200. But the gains are in infrastructure and engineering. The losses are concentrated in office-based, entry-level roles. The distribution matters as much as the net number.
Every previous automation wave targeted manual, physical, or routine tasks. AI is the first technology to target cognitive, creative, and analytical work. This means the historical pattern — displaced manual workers moving into thinking roles — may not hold. Where do displaced knowledge workers go when AI can do the knowledge work?
📝 Freelancers — The Canary in the Coal Mine
Freelancers are the canary in the coal mine for AI displacement. They lack the institutional buffers — retraining programmes, severance, notice periods — that shield full-time employees. Harvard research shows the impact is already measurable across multiple freelance categories.
Harvard researchers tracked freelance job postings across Upwork, Fiverr, and similar platforms since ChatGPT’s launch. The declines are steep and concentrated in the categories where AI tools are strongest: writing, basic software development, and graphic design. Junior-level positions were hit hardest.
The corporate spending data tells the same story from the demand side. Ramp’s analysis of business spending patterns shows freelancer/contractor spend collapsing by 33% while AI model spending rose sharply. Companies are not just hiring fewer freelancers — they are actively substituting AI tools for freelance work.
Categories Hit Hardest
- Copywriting and content writing (−33%)
- Software development (−21%)
- Graphic design (−17%)
- Junior-level positions (steepest drops)
- Routine, templated, or bulk work orders
Categories Holding Steady
- Senior/specialist strategy consulting
- Complex systems architecture
- Client-facing advisory work
- Regulated compliance writing
- Physical/on-site services
The pattern within freelancing mirrors the broader employment picture: work that can be fully specified in a text prompt is disappearing. Work that requires context, relationships, physical presence, or licensed expertise is holding. The dividing line runs through every freelance category, not between them.
| Finding | Value | Source |
|---|---|---|
| Freelance writing job postings decline since ChatGPT launch (US) | -30% | Harvard / Imperial College London (2024) |
| Freelance software development postings decline (US) | -21% | Harvard / Imperial College London (2024) |
| Freelance graphic design postings decline (US) | -17% | Harvard / Imperial College London (2024) |
| Drop in junior-level freelance positions (US) | -7.7% | Harvard Economics (Lichtinger & Hosseini Maasoum, 2025) |
| Decline in corporate freelancer spending, 2025 (US) | 0.66% → 0.14% | Ramp “Payrolls to Prompts” (Feb 2026) |
| Increase in corporate AI model spending, 2025 (US) | 0% → 2.85% | Ramp “Payrolls to Prompts” (Feb 2026) |
Freelancers lack the institutional buffers that slow displacement for full-time employees: no retraining programmes, no severance, no notice periods, no internal redeployment. When AI can do the work, the freelance contract simply does not get renewed. What is happening to freelancers now is a preview of what may happen to full-time roles in 12–24 months.
💰 The Economic Domino Effect
Mass AI unemployment doesn’t just mean people without jobs. It means missed loan payments, cascading defaults, shrinking consumer demand, and a self-reinforcing downturn. JP Morgan calls this the “strain before the boom” scenario: AI suppresses demand through displacement before productivity gains reach consumers.
Source: Citadel Securities; Fortune/Citrini analysis; BLS
JP Morgan Private Bank calls this the “strain before the boom” scenario: AI suppresses demand through job displacement before productivity gains reach consumers. The OECD has flagged an AI bubble burst as a key 2026 risk. If AI capex — which powered 92% of GDP growth in H1 2025 — slows, the economy loses its primary growth engine.
Goldman Sachs projects a 15% labour productivity boost from AI adoption, but this is a 10–15 year timeline. The displacement comes first. The productivity gains come later. The policy question is whether the transition period is 2 years (Goldman’s optimistic model) or 5–10 years (the pessimistic model).
The Domino Chain: Displacement to Contraction
Displacement — AI replaces tasks, companies cut headcount or freeze hiring.
Income loss — Displaced workers reduce spending. Freelancers lose contracts first, followed by full-time employees.
Demand contraction — Consumer spending falls in affected regions and sectors.
Secondary effects — Local businesses serving displaced workers also contract. Loan defaults rise. Housing markets soften.
Productivity gains — Eventually, AI-driven productivity growth creates new demand and new roles. Goldman estimates 2 years. Pessimists say 5–10.
The economic risk depends on how long steps 1–4 run before step 5 kicks in.
| Finding | Value | Source |
|---|---|---|
| AI capital expenditure as share of US GDP growth (H1 2025) | 92% | Citadel Securities / Fortune |
| Total AI capital expenditure, 2% of GDP (US) | $650B | Citadel Securities |
| Net US jobs created by AI in 2024 (gains minus losses) | +107,200 | ITIF (Dec 2025) |
| US labour productivity growth (2025) | 2.2% | BLS |
| Goldman Sachs: projected AI labour productivity boost (Global) | ~15% | Goldman Sachs (Aug 2025) |
| Goldman Sachs: displacement resolution timeline (Global) | 2 years | Goldman Sachs (Aug 2025) |
Displacement is immediate. Productivity gains are gradual. This timing mismatch is the core economic risk. Workers lose income now. New jobs are created later. If the gap is short (Goldman’s 2-year model), the economy adapts. If it is long (5–10 years), the cascading effects — missed payments, defaults, reduced spending — can become self-reinforcing.
🛡️ What Protects Workers From AI Displacement
The data reveals four traits that consistently protect roles from AI displacement: physical presence in unpredictable environments, regulatory licensing, human judgement under uncertainty, and interpersonal trust. Roles with these traits cluster in the GREEN zone regardless of industry.
Physical Presence
Roles requiring hands-on work in unpredictable environments score highest. AI cannot rewire a building, treat a patient in a hospital bed, or navigate a construction site. This is the strongest single protection factor.
Regulatory Licensing
Professional licensing creates a structural barrier. Even if AI can perform the task, law and regulation require a licensed human to sign off. This protection persists until legislation changes — which moves slowly.
Human Judgement Under Uncertainty
Roles that require navigating ambiguity, making high-stakes decisions with incomplete data, and adapting in real time resist automation. AI excels at pattern matching, not novel uncertainty.
Interpersonal Trust
Roles built on personal relationships — therapists, social workers, senior advisors — require a level of human trust that AI has not earned. Trust is a human-to-human construct.
The 15 most AI-resistant roles in our database (highest scores):
| # | Role | Score |
|---|---|---|
| 1 | Electrical Power-Line Installer and Repairer (Mid-Level) | 91.6 /100 |
| 2 | Signalling Tester In Charge / STIC (Mid-Level) | 87.7 /100 |
| 3 | Model Alignment Researcher (Mid-Level) | 86.1 /100 |
| 4 | AI Safety Researcher (Mid-Senior) | 85.2 /100 |
| 5 | Foster Carer (Mid-Level) | 84.5 /100 |
| 6 | Leadworker (Mid-Level) | 83.7 /100 |
| 7 | Heat Pump Installer (Mid-Level) | 83.5 /100 |
| 8 | Trauma Surgeon (Mid-to-Senior) | 83.2 /100 |
| 9 | CCS Engineer (Control Command & Signalling) (Mid-Level) | 83.2 /100 |
| 10 | Chief Information Security Officer (CISO) (Senior/Executive) | 83.0 /100 |
| 11 | Electrician (Journey-Level) | 82.9 /100 |
| 12 | Intimacy Coordinator (Mid-Level) | 82.6 /100 |
| 13 | Master Leather Craftsman (Mid-to-Senior) | 82.4 /100 |
| 14 | Registered Nurse (Clinical/Bedside) | 82.2 /100 |
| 15 | Complex Family Planning Specialist (Mid-to-Senior) | 82.0 /100 |
For career guidance based on this data:
- AI-Proof Career Guide — practical steps for career resilience
- Jobs Safe From AI — the GREEN zone roles with the strongest defences
- Most AI-Proof Jobs — the highest-scoring roles in our database
The strongest AI defences combine multiple traits. A licensed nurse practitioner working in a hospital has physical presence + licensing + human judgement + interpersonal trust. A remote data analyst has none of these. Protection is cumulative — roles with two or more traits consistently score above the GREEN zone threshold.
🚨 The Safety Net Isn’t Ready
If AI unemployment accelerates, existing systems cannot handle it. 75% of unemployed Americans never apply for unemployment insurance. The system’s framework has not been updated since the New Deal. Federal taxes funding it haven’t changed since the 1980s.
The UI Gap
Nearly 75% of unemployed Americans do not even apply for unemployment insurance. Of those who do, only 55% receive benefits. The system’s framework has not been updated since the New Deal. Federal taxes funding it have not changed since the 1980s. Some states cut benefits to 12 weeks.
Source: BLS; Fortune, March 2026
The Vulnerable 6.1 Million
Brookings identifies 6.1 million US workers with high AI exposure and low adaptive capacity — limited savings, low skill transferability, narrow reemployment prospects. 86% are in clerical and administrative roles concentrated in smaller metro areas.
Source: Brookings Institution, 2026
The UBI Experiments
72 UBI pilots across 26 US states since 2020, distributing $335 million to ~30,000 people. Cook County (Chicago) became the first US government body to make guaranteed income permanent — $500/month to 3,200 households. No pilots report employment declines.
Source: Economic Security Project; Business Insider
The Hinton Position
Geoffrey Hinton, the Nobel Prize-winning AI pioneer, dismissed UBI as a solution: it cannot address the loss of dignity and purpose that comes with losing work. The policy debate is still catching up to the technology.
Source: Bloomberg TV, Nov 2025
| Finding | Value | Source |
|---|---|---|
| Unemployed who never apply for unemployment insurance (US) | ~75% | BLS (2023 survey) |
| Of those who apply for UI, share who receive benefits (US) | 55% | BLS / Fortune (2023 survey) |
| WEF: workers needing reskilling by 2027 (Global) | 60% | World Economic Forum |
| Goldman Sachs: workers needing significant upskilling by 2030 (Global) | 40%+ | Goldman Sachs (Aug 2025) |
| UBI pilot programmes launched since 2020 (US) | 72 (across 26 states) | Economic Security Project |
| World Bank: displacement risk in developing economies (Global) | Up to 30% | World Bank / Industry synthesis |
US unemployment insurance was designed for cyclical recessions — temporary dips in demand that resolve when the economy recovers. AI displacement is structural: the roles may not come back. A system designed for 6-month disruptions cannot handle permanent occupational shifts.
🚦 The Reskilling Challenge
The WEF estimates 60% of workers need reskilling by 2027. Goldman Sachs says over 40% need significant upskilling by 2030. Yet only 1% of companies report “mature” AI deployment. The reskilling demand hasn’t peaked — it’s barely started.
The scale of the reskilling challenge is without precedent. 60% of workers needing new skills by 2027 — just 1 year away — while only 1% of companies have achieved mature AI deployment. The reskilling demand is running far ahead of the adoption that would justify it, creating a paradox: workers are being told to reskill for an AI-transformed workplace that most employers have not yet built.
What to Reskill Toward
- Roles requiring physical presence + licensing
- AI-adjacent roles (AI auditing, AI training, prompt engineering)
- Skilled trades facing structural shortages
- Healthcare and social care (ageing population demand)
- Cybersecurity (persistent talent deficit)
What to Reskill Away From
- Purely digital, repetitive knowledge work
- Data entry and basic administrative tasks
- Routine content generation (copywriting, basic design)
- Manual bookkeeping and tax preparation
- Roles with zero physical or regulatory barriers
For specific career pathways, see our AI-Proof Career Guide — practical steps mapped to our scoring data.
The WEF says 60% of workers need reskilling by 2027. But most workers do not know what to reskill for because the technology is changing faster than training curricula can adapt. The safest reskilling targets are not specific AI tools (which become obsolete quickly) but the human traits that AI cannot replicate: physical dexterity, licensed judgement, interpersonal trust.
📖 Sources
77 externally sourced statistics from 59+ institutions, linked to original sources.
- US unemployment rate (early 2026): 4.28% — BLS / Citadel Securities
- AI-attributed US job losses in 2025: 55,000 — Challenger, Gray & Christmas
- AI share of total US job losses (2025): 4.5% — Challenger, Gray & Christmas
- Employment in high-AI-exposure occupations, mid-2024 to late 2025 (US): +1.7% — Yale Budget Lab (Jan 2026)
- St. Louis Fed: AI exposure ↔ unemployment correlation (US): 0.47 — Federal Reserve Bank of St. Louis (Aug 2025)
- Companies reducing hiring in anticipation of AI (US): 29% — HBR (Jan 2026)
- Companies planning to replace workers with AI by year-end 2026 (US): 37% — Resume.org (1,000 US leaders)
- Organizations that made large AI-driven workforce reductions (US): 2% — HBR (Jan 2026)
- Employers expecting AI headcount cuts in 2026 (US): 1 in 6 — Industry surveys
- US workers using GenAI daily in their job: 12% — St. Louis Fed
- Goldman Sachs: projected temporary US unemployment rise: +0.5pp — Goldman Sachs (Aug 2025)
- Goldman Sachs: US workforce displacement range: 6–7% (range 3–14%) — Goldman Sachs (Aug 2025)
- Goldman Sachs: global full-time jobs exposed to generative AI: 300 million — Goldman Sachs
- Anthropic CEO: possible US unemployment from AI: 10–20% — Dario Amodei (May 2025)
- JP Morgan: US displacement in 1–3 years: 3–6% — JP Morgan Private Bank
- Geoffrey Hinton: massive unemployment likelihood (Global): "Very likely" — Geoffrey Hinton (Nobel Prize, Bloomberg TV Nov 2025)
- Federal Reserve: projected unemployment end 2026 (US): 4.4% — Federal Reserve
- IMF: global jobs exposed to AI (Global): 40% — International Monetary Fund (2024)
- IMF: jobs exposed in advanced economies (Global): 60% — International Monetary Fund (2024)
- WEF: global jobs displaced by 2030 (Global): 92 million — World Economic Forum (2025)
- WEF: global jobs created by 2030 (Global): 170 million — World Economic Forum (2025)
- McKinsey: US occupational transitions needed by 2030: 12 million — McKinsey Global Institute
- MIT: US workforce whose tasks AI can already perform: ~12% — MIT (Nov 2025)
- OECD: jobs in high-exposure occupations, 50%+ automatable (Global): 27% — OECD Employment Outlook 2023
- PwC: jobs automatable by mid-2030s (Global): Up to 30% — PwC
- McKinsey: US work hours automatable with AI: 57% — McKinsey Global Institute (2025)
- Companies already replacing workers with AI (US): 30% — Resume.org (1,000 US leaders)
- Companies using AI as a scapegoat for layoffs (US): 59% — Resume.org (1,000 hiring managers)
- Employers planning workforce reduction due to AI (Global): 40% — World Economic Forum
- BT Group: AI-attributed job cuts (UK): 55,000 total / 10,000 AI-replaced — BT Group (2023)
- January 2026 US job cuts: 108,435 — Challenger, Gray & Christmas
- AI layoffs that were anticipatory (US): 77% — HBR (Jan 2026)
- Companies with mature AI deployment (Global): 1% — McKinsey State of AI (2024)
- Organizations using AI in one function (Global): 78% — McKinsey (2025)
- Stanford: employment decline ages 22–25 in AI-exposed jobs (US): -16% — Stanford DEL (Brynjolfsson et al., 2025)
- Big Tech new-grad hiring cut, 2023–2024 (US): -25% — Goldman Sachs (2025)
- IMF: women’s employment vulnerability to AI (Global): 9.6% — IMF (2024)
- Workers with high AI exposure + low adaptive capacity (US): 6.1 million — Brookings Institution (2026)
- Global employee fear of AI job loss (2024 → 2026): 28% → 40% — Mercer (12,000 respondents)
- Pew: US workers worried about AI in the workplace: 52% — Pew Research (Oct 2024)
- Workers saying AI job elimination “not at all likely” (US): 50% (down from 60% in 2023) — Gallup (2025)
- New college graduate unemployment, early 2026 (US): ~10% — Goldman Sachs / Industry data
- US youth unemployment ages 20–24 (Sep 2025): 9.5% — BLS / Fortune
- Drop in young workers finding jobs in AI-exposed roles (US): -14% — Anthropic Research (2025)
- Anthropic CEO: entry-level white collar displacement (US): 50% within 1–5 years — Dario Amodei (May 2025)
- Harvard: freelance writing decline (US): -30% — Harvard / Imperial College London (2024)
- Harvard: freelance dev decline (US): -21% — Harvard / Imperial College London (2024)
- Harvard: freelance design decline (US): -17% — Harvard / Imperial College London (2024)
- Harvard: junior freelance positions decline (US): -7.7% — Harvard Economics (Lichtinger & Hosseini Maasoum, 2025)
- Ramp: freelancer spend collapse, 2025 (US): 0.66% → 0.14% — Ramp “Payrolls to Prompts” (Feb 2026)
- Ramp: AI model spend rise, 2025 (US): 0% → 2.85% — Ramp “Payrolls to Prompts” (Feb 2026)
- ILO: global unemployment rate: 5.0% — ILO World Employment & Social Outlook 2025
- ILO: total global employment: 3.44B — ILO World Employment & Social Outlook 2025
- ILO: global youth unemployment: 13.0% — ILO World Employment & Social Outlook 2025
- UK unemployment rate: 4.4% — ONS Labour Market Overview
- German unemployment rate: 6.0% — Destatis / Bundesagentur für Arbeit
- EU unemployment rate: 5.9% — Eurostat Unemployment Statistics
- EU youth unemployment, under 25 (EU): 14.3% — Eurostat Youth Employment Statistics
- Canadian unemployment rate: 6.7% — Statistics Canada Labour Force Survey
- Australian unemployment rate: 4.1% — ABS Labour Force
- India unemployment rate (India): 3.2% — PLFS Annual Report 2024-25
- OECD average unemployment, Jul 2025 (Global): 4.9% — OECD
- UK workers fearing AI job loss (UK): 25%+ — Randstad Survey (Jan 2026)
- UK AI adoption by businesses, 2023 (UK): ~15% — ONS
- Work tasks automatable by 2030 (Global): 34% — Tenet / Employer Surveys
- AI capex as share of US GDP growth (H1 2025): 92% — Citadel Securities / Fortune
- AI capital expenditure, 2% of GDP (US): $650B — Citadel Securities
- AI jobs created in 2024, net of losses (US): +107,200 — ITIF (Dec 2025)
- US labour productivity growth (2025): 2.2% — BLS
- Goldman Sachs: AI productivity boost (Global): ~15% — Goldman Sachs (Aug 2025)
- Goldman Sachs: displacement resolves within 2 years (Global): 2 years — Goldman Sachs (Aug 2025)
- Unemployed who don’t apply for UI benefits (US): ~75% — BLS (2023 survey)
- UI applicants who receive benefits (US): 55% — BLS / Fortune (2023 survey)
- WEF: workers needing reskilling by 2027 (Global): 60% — World Economic Forum
- Goldman Sachs: workers needing upskilling by 2030 (Global): 40%+ — Goldman Sachs (Aug 2025)
- US UBI pilots since 2020: 72 (across 26 states) — Economic Security Project
- World Bank: displacement in developing economies (Global): Up to 30% — World Bank / Industry synthesis
✅ The Bottom Line
AI is not causing mass unemployment today. But it is causing measurable displacement in specific roles, sectors, and demographics. 516 roles covering 44.3M US workers sit in the RED zone. Young workers, women in clerical roles, and freelancers are bearing the earliest impact. The gap between expert predictions (0.5pp vs 20%) reflects genuine uncertainty about how fast AI capabilities will advance.
The data does not support either extreme. Mass unemployment is not imminent — but structural displacement is already underway. The workers most at risk share a common trait: their work lives entirely in software, in a digital-first environment where AI is strongest.
The workers most protected share the opposite trait: their work requires physical presence, licensed human judgement, or interpersonal trust that AI has not replicated.
If you are reading this worried about your career: check your role’s individual assessment. The answer is not “all jobs are at risk” or “no jobs are at risk.” It depends on what you do, how you do it, and whether AI can replicate the specific tasks that make up your working day.
What's your AI risk score?
We're building a free tool that analyses your career against millions of data points and gives you a personal risk score with transition paths. We'll only build it if there's demand.
No spam. We'll only email you if we build it.
The AI-Proof Career Guide
We've found clear patterns in the data about what actually protects careers from disruption. We'll publish it free — but only if people want it.
No spam. We'll only email you if we write it.
About This Data
AIJRI scores are generated using the AI Job Resistance Index methodology v3, a composite scoring framework evaluating each role across resistance, evidence, barriers, protective principles, and AI growth correlation. Scores range from 0 (no resistance) to 100 (maximum resistance). Our database currently covers 3649 roles mapped to US BLS employment data, representing approximately 100% of the 168.7M US workforce.
Zone thresholds: GREEN (≥48) = AI-resistant, YELLOW (33–47) = transforming, RED (<33) = high displacement risk. RED Imminent (<20) = AI can already perform most core tasks.
UK and Global workforce figures are proportional estimates based on our assessed US zone distribution applied to ONS (UK) and ILO (Global) total employment data. Actual exposure may differ due to industry composition and regulatory environment.
External statistics (77 data points from 59+ institutions) are cited with their original sources and linked where available. Our AIJRI data updates dynamically as new assessments are added. External figures are editorial context — they do not power any calculations on this page.
Related: AI and Job Loss Statistics · Jobs Most at Risk From AI · What Jobs Will AI Replace First? · Is AI Replacing Entry Level Jobs?
About the Authors
Nathan House
AI and cybersecurity expert with 30 years of hands-on experience. Nathan founded StationX (500,000+ students) and built JobZone Risk to ensure people invest their career development in the right direction.
StationX HAL
Custom AI infrastructure built by Nathan House for StationX. HAL co-develops JobZone Risk end-to-end: the scoring methodology, the assessment pipeline, every role assessment, and the statistical analysis that powers these articles — all directed by Nathan.