AI Job Losses by Industry 2026
By Axis Intelligence Research
Co-author: David Park | Last updated: August 19, 2026 | License: CC BY 4.0
American employers attributed 112,713 announced job cuts to artificial intelligence in the first seven months of 2026 — more than double the 54,836 recorded in all of 2025. Axis Intelligence Research finds those cuts are not spread across the economy. Measured against sector payrolls, technology absorbed 4,157 announced cuts per 100,000 workers; health care absorbed 144.
Quick Answer
AI-attributed job losses in 2026 are concentrated, not general. Axis Intelligence Research’s AI Labor Pressure Index (AILP) scores technology at 93.1 out of 100 and health care at 15.6 — a six-fold spread across US sectors. Announced cuts tied to AI reached 112,713 through July 2026, equal to 0.071% of the 158.9 million nonfarm payroll jobs the Bureau of Labor Statistics counted that month.
Key Findings
- According to Challenger, Gray & Christmas, US employers cited artificial intelligence in 112,713 announced job cuts between January and July 2026, or roughly 24% of all cuts announced in that period.
- Axis Intelligence Research finds technology recorded 4,157 announced job cuts per 100,000 employees in 2026 through July — 28.9 times the rate recorded in health care.
- Axis Intelligence Research’s AI Labor Pressure Index ranks technology first at 93.1 and health care last at 15.6 among eight major US sectors, as of August 19, 2026.
- The Stanford Digital Economy Lab reported on August 12, 2026 that employment for workers aged 22–25 in highly AI-exposed occupations sits about 19% below its less-exposed benchmark, up from 15% at the July 2025 data vintage.
- Axis Intelligence Research estimates the 112,713 AI-attributed cuts announced in 2026 through July represent roughly $7.56 billion in annualised payroll, derived from BLS average weekly earnings of $1,290.37.
How Many Jobs Has AI Cut in 2026?
The count depends entirely on who is doing the counting, and the gap between methods is the whole story.
Challenger, Gray & Christmas tracks the reason employers give when they announce layoffs. In its July 2026 report, the firm recorded 10,970 AI-attributed cuts for the month, the fifth consecutive month AI led all stated reasons, bringing the 2026 total to 112,713. Since Challenger began tracking AI as a distinct category in 2023, the cumulative figure is 184,538.
Set that against the denominator. The Bureau of Labor Statistics counted 158,858,000 nonfarm payroll jobs in July 2026. The 2026 AI-attributed total is 0.071% of that base — seven hundredths of one percent.
| Measure | Value | Period | Source |
|---|---|---|---|
| AI-attributed announced cuts | 112,713 | Jan–Jul 2026 | Challenger, Gray & Christmas |
| AI-attributed cuts, full year | 54,836 | 2025 | Challenger, Gray & Christmas |
| AI-attributed cuts, cumulative | 184,538 | 2023–Jul 2026 | Challenger, Gray & Christmas |
| All announced cuts | 477,033 | Jan–Jul 2026 | Challenger, Gray & Christmas |
| Total nonfarm payrolls | 158,858,000 | July 2026 | BLS Table B-1 |
| AI cuts as share of payrolls | 0.071% | Jul 2026 | Axis Intelligence Research calculation |
Two things are true at once and the discourse keeps collapsing them. AI-attributed cuts more than doubled year over year — that is a real acceleration. And they remain a rounding error against total employment. A worker in an affected requisition does not care about the denominator, and a policymaker cannot ignore it.
There is also a measurement problem Challenger names openly. The firm tracks a separate category, “Technological Update (possibly AI),” for announcements where new technology is blamed and AI is implied but not stated — 20,219 such cuts in 2025. And Andy Challenger has pointed out that naming AI in a layoff announcement can play well with investors while playing badly with staff, which means the reason companies state is a communications decision as much as an operational one. Employer self-attribution is a useful signal. It is not a measurement of causation, and nobody should treat it as one.
Which Industries Are Losing the Most Jobs to AI?
Raw cut totals rank industries by size more than by exposure. A sector with 24 million workers will out-cut a sector with 1.5 million workers almost regardless of what technology is doing to it. So the useful question is cuts per worker.
Announced Job Cuts per 100,000 Employees, 2026
| Industry | Announced cuts (Jan–Jul 2026) | Employment, Jul 2026 | Cuts per 100,000 | Source |
|---|---|---|---|---|
| Technology | 149,023 | 3,584,900 | 4,157.0 | Challenger; bls.gov |
| Transportation & warehousing | 58,076 | 6,596,100 | 880.5 | Challenger; bls.gov |
| Media | 4,428 | 1,561,800 | 283.5 | Challenger; bls.gov |
| Financial | 18,626 | 9,090,000 | 204.9 | Challenger; bls.gov |
| Health care | 34,426 | 23,912,200 | 144.0 | Challenger; bls.gov |
| Professional & business services | 23,942 | 22,518,000 | 106.3 | Challenger; bls.gov |
| Government | 20,752 | 23,270,000 | 89.2 | Challenger; bls.gov |
| Retail | 12,946 | 15,440,900 | 83.8 | Challenger; bls.gov |
Employment figures are seasonally adjusted, from BLS Current Employment Statistics Table B-1. Sector mappings between Challenger’s industry taxonomy and BLS NAICS supersectors are documented in the Methodology.
Technology carries 4.7 times the cut density of transportation and warehousing, the next-ranked sector, and 28.9 times health care’s. Challenger’s own reporting supports the concentration: technology alone accounted for 31% of all announced cuts in 2026 through July, and Andy Challenger has described the sector as “the center of gravity” for the year’s reductions.
Health care deserves a closer look, because it is where the gap between narrative and data is widest. Challenger’s data contains exactly one health care cut attributed to AI — 39 positions at a California telehealth provider in October 2025. Health care’s 34,426 announced cuts in 2026 are real, and up 6% year over year, but they are hospital finance and product-line decisions, not substitution events. Meanwhile the sector added 552,100 jobs over the year to July 2026, the largest absolute gain of any industry BLS tracks.
David Park: The sector that dominates the “AI is coming for your job” headlines is the one BLS shows hiring hardest. That is not a paradox, it’s a category error — health care demand is set by demographics, not by software procurement, and no model has yet found a way to reduce the number of 78-year-olds who need care. If you are choosing a field in 2026 and you have been reading that AI is eating employment, the published requisition counts point somewhere different than the headlines do.
What Is the AI Labor Pressure Index (AILP)?
Cut density alone is a snapshot of announcements. It says nothing about whether a sector’s payroll is actually shrinking, and nothing about whether AI is even deployed there. Axis Intelligence Research built the AI Labor Pressure Index (AILP) to combine all three.
AILP measures the degree of AI-related labour-market pressure on a US industry, on a 0–100 scale, as of a stated date. It is not a forecast and not a causal estimate. It is a composite of three independently sourced, independently verifiable components.
AILP Formula and Weights
AILP = (0.40 × C1) + (0.35 × C2) + (0.25 × C3)
C1 Cut density Announced job cuts per 100,000 employees,
(40%) Jan–Jul 2026. Min-max normalised 0–100
across the eight sectors.
Sources: Challenger, Gray & Christmas; BLS CES Table B-1.
C2 Employment Year-over-year change in sector payroll employment,
trajectory July 2025 → July 2026, seasonally adjusted.
(35%) Inverted and min-max normalised, so contraction
scores higher than growth.
Source: BLS CES Table B-1.
C3 Adoption Ordinal tier from the 2026 AI supplement to the
exposure Census Bureau's Business Trends and Outlook Survey:
(25%) 100 = sector named among highest AI-use sectors
(Information, Professional Services, Finance)
60 = knowledge-intensive, not named
25 = predominantly physical or in-person work
Source: US Census Bureau, CES-WP-26-25.
Weights favour C1 because announced cuts are the most direct observable, and hold C3 lowest because an ordinal tier carries less information than a continuous measure. Any competent analyst can recompute every score from the CSV that ships with this page.
AILP Scores by Industry — Baseline Reading, August 19, 2026
| Rank | Industry | C1 Cut density | C2 Trajectory | C3 Adoption | AILP |
|---|---|---|---|---|---|
| 1 | Technology | 100.0 | 80.4 | 100 | 93.1 |
| 2 | Media | 4.9 | 100.0 | 100 | 62.0 |
| 3 | Financial | 3.0 | 69.8 | 100 | 50.6 |
| 4 | Government | 0.1 | 71.6 | 60 | 40.1 |
| 5 | Professional & business services | 0.6 | 35.8 | 100 | 37.8 |
| 6 | Transportation & warehousing | 19.6 | 64.2 | 25 | 36.5 |
| 7 | Retail | 0.0 | 44.4 | 25 | 21.8 |
| 8 | Health care | 1.5 | 0.0 | 60 | 15.6 |
Component scores are normalised within this eight-sector set; a score of 0 means lowest in the set, not zero pressure. This is the baseline reading.
Three results are worth sitting with.
Technology scores 93.1 because it is the only sector where all three components point the same direction: the highest cut density by a factor of nearly five, a payroll that shrank 1.79% over the year, and top-tier AI deployment. Nothing else in the US economy currently looks like that.
Media scores 62.0 on almost no cut volume — 4,428 announced cuts is small — but it posted the steepest payroll contraction of any sector measured here, down 2.80% year over year, alongside top-tier adoption. Publishing, broadcasting and motion picture employment fell by a combined 45,000 jobs. Media’s pressure shows up in the payroll, not the press release.
Professional and business services scores only 37.8 despite top-tier AI adoption, because its payroll grew by 115,000 jobs over the year. The sector that Census identifies as among the heaviest AI users is not shedding workers in aggregate. That is the single most important corrective to the “AI is replacing white-collar work” framing, and it comes from federal payroll data, not opinion.
David Park: I keep meeting people who have read that professional services is next and are quietly panicking about a career they have twenty years left in. The BLS number for that sector is plus 115,000 over the year. That is not a promise, and the composition inside the total is changing in ways the headline hides — which is the next section. But the aggregate is growing, and anyone telling you otherwise is not reading the release.
What Is the Estimated Marginal Cost of AI-Attributed Job Cuts?
Every displacement decision is a price comparison. The employer is weighing an annual wage bill against a software bill, and the cuts that get announced are the ones where the comparison came out a particular way. Putting a number on the wage side makes the trade explicit.
Axis Intelligence Research estimates the annualised payroll represented by 2026’s AI-attributed cuts as follows.
Annualised earnings per worker
= average hourly earnings × average weekly hours × 52
= $37.62 × 34.3 × 52
= $1,290.37 per week × 52
= $67,099
Marginal payroll exposure
= 112,713 AI-attributed cuts × $67,099
= $7.56 billion (annualised)
Cumulative, 2023 – July 2026
= 184,538 cuts × $67,099
= $12.38 billion (annualised)
Inputs are BLS Employment Situation figures for July 2026: average hourly earnings of $37.62 for all employees on private nonfarm payrolls and an average workweek of 34.3 hours. The derived weekly figure of $1,290.37 matches BLS’s own published average weekly earnings exactly, which is a useful confirmation that the inputs are being combined correctly.
| Input | Value | As of | Source |
|---|---|---|---|
| Average hourly earnings, private | $37.62 | July 2026 | BLS Employment Situation |
| Average weekly hours, private | 34.3 | July 2026 | bls.gov |
| Average weekly earnings, private | $1,290.37 | July 2026 | bls.gov |
| Annualised earnings per worker | $67,099 | July 2026 | Axis Intelligence Research calculation |
| Marginal payroll exposure, 2026 YTD | $7.56B | Jul 2026 | Axis Intelligence Research calculation |
This is an estimate of gross payroll represented, not of realised savings. It applies an economy-wide average wage to cuts that skew toward technology, where pay runs above average, so it is more likely conservative than inflated. It also excludes benefits, severance and the cost of the systems replacing the work. Read it as an order of magnitude for the wage side of the trade: under $8 billion a year, against a US private payroll measured in trillions.
David Park: Seven and a half billion sounds enormous until you set it beside what the same companies are spending on the infrastructure meant to do the work. The substitution is not obviously cheap yet. It is being priced as though it will be.
Is AI Displacement Concentrated in Tech, or Spreading?
Here the two best datasets disagree, and the disagreement is more informative than either alone.
Challenger’s position is explicit: AI-related cutting has been limited outside technology. Their by-industry evidence supports it — a single AI-attributed health care cut in the entire dataset.
Federal administrative data says something different. In an April 2026 Census Bureau working paper, economist Lee Tucker analysed matched employer-employee records and found that early-career employment in the most AI-exposed quintile of industry-state cells fell 12% over the ten quarters following ChatGPT’s release, driven by a 9% drop in hiring. The finding that matters for this page is the one about breadth: although the most-exposed quintile is dominated by a handful of sectors, Tucker reports that the association between higher AI exposure and reduced early-career hiring appears across most sectors of the economy.
The Stanford Digital Economy Lab reached a compatible conclusion on August 12, 2026. Using ADP payroll records, Brynjolfsson, Chandar and Chen reported no widespread economy-wide displacement, but a gap for workers aged 22–25 in highly AI-exposed occupations that has widened to roughly 19%, from 15% at the July 2025 vintage. Their fourth stated fact is the mechanism: adjustment is running through reduced hiring of young workers rather than increased separations.
That reconciles the disagreement. Challenger counts announcements; announcements are separations. Census and Stanford count hires; a job that is never posted generates no announcement, no WARN notice and no Challenger row. If AI displacement operates primarily by shrinking entry-level intake rather than by cutting incumbents, then a layoff-announcement tracker will register it only where firms are also actively reducing headcount — which in 2026 means technology.
| Finding | Value | Source | Method |
|---|---|---|---|
| Early-career employment decline, most-exposed quintile | −12% | US Census Bureau (Tucker, 2026) | Matched employer-employee records |
| Early-career hiring decline, most-exposed quintile | −9% | US Census Bureau (Tucker, 2026) | Matched employer-employee records |
| Employment gap, ages 22–25, high-exposure occupations | ~19% | Stanford Digital Economy Lab | ADP payroll data |
| Same gap, July 2025 vintage | ~15% | Stanford Digital Economy Lab | ADP payroll data |
| Firms reporting AI-related employment decreases | 2% | US Census Bureau, BTOS AI supplement | Nationally representative survey |
Axis Intelligence Research reads the Stanford result with one caveat the authors themselves put in writing: the estimated gaps are larger in the ADP sample than in national survey benchmarks, and they narrow when education is controlled for. The lab describes its own findings as descriptive rather than causal. Publications that have quoted the 19% figure as settled proof of AI displacement have quoted it more confidently than the researchers who produced it.
How Widely Is AI Actually Deployed by Sector?
Displacement requires deployment, and deployment is thinner than the discourse implies.
The 2026 AI supplement to the Census Bureau’s Business Trends and Outlook Survey surveyed firms between November 2025 and January 2026. Across the economy, 18% of firms used AI in a business function — 32% on an employment-weighted basis, since larger firms adopt more. Among very large firms in information, professional services and finance, use rates reached 50–60%, or 60–70% employment-weighted.
Two numbers from that survey belong in every discussion of AI and jobs. Among firms using AI, 66% use it solely to augment tasks. And AI-related employment decreases were reported by 2% of firms.
| Metric | Value | Reference period | Source |
|---|---|---|---|
| Firms using AI in a business function | 18% | Nov 2025–Jan 2026 | census.gov |
| Same, employment-weighted | 32% | Nov 2025–Jan 2026 | census.gov |
| Expected adoption within six months | 22% | Nov 2025–Jan 2026 | census.gov |
| Use rate, very large firms in Information/Professional/Finance | 50–60% | Nov 2025–Jan 2026 | census.gov |
| Firms using AI solely to augment tasks | 66% | Nov 2025–Jan 2026 | census.gov |
| Firms reporting AI-related employment decreases | 2% | Nov 2025–Jan 2026 | census.gov |
| Users integrating AI in three or fewer functions | 57% | Nov 2025–Jan 2026 | census.gov |
The Census authors also found that functional breadth and operational investment are positively associated with employment decreases, while worker-task use shows no significant link to headcount reduction once those are accounted for. Restated for the people making career decisions: your colleagues using a chatbot is not the displacement signal. Your employer rebuilding a business function around AI is.
David Park: Two percent. That is the share of US firms reporting AI-related headcount reductions in nationally representative federal survey data, during the period when the technology was supposedly ending white-collar work. Hold that number next to the projections and notice which one has a sample frame behind it.
Which Workers Are Most Exposed, and Why Age Matters More Than Industry
The strongest finding across all three research sources is not about industry at all. It is about tenure.
Stanford’s revision introduces a distinction that explains the age gradient better than sector does: codified versus tacit knowledge. Employment fell among young workers in occupations built on formal, documented, teachable knowledge — the kind that has already been written down and can therefore be reproduced from text. Employment rose among experienced workers in occupations built on knowledge acquired through practice and mentorship. The lab also reports that women face greater AI exposure on average.
That framing does more work than any industry ranking. It explains why software development and customer service show up together despite sitting in different sectors. It explains why the gap widened through mid-2026, well after interest rates peaked, which is the timing objection the interest-rate explanation cannot answer. And it explains why a 24-year-old and a 44-year-old in the same job title at the same firm are experiencing different labour markets.
David Park: The fear I hear most from people at 11pm is that their degree has been rendered worthless. The data says something narrower and more actionable: the entry to the ladder is what has narrowed, and it narrowed by hiring freeze rather than by firing. BLS put the unemployment rate for workers with a bachelor’s degree or higher at 2.7% in July 2026, unchanged over the month and over the year. If you are already on the ladder, the published numbers are not describing your situation. If you are trying to get on it, they are.
Methodology
Scope. This analysis covers the United States, January through July 2026, with employment levels as of July 2026. Eight sectors are included; sectors were selected where Challenger’s industry taxonomy could be mapped to BLS NAICS supersectors without double-counting.
Data collection. All figures were retrieved from primary sources between August 17 and August 19, 2026. Employment levels come from BLS Current Employment Statistics Table B-1, seasonally adjusted, July 2026 preliminary, benchmarked to March 2025 levels. Announced job cuts come from Challenger, Gray & Christmas monthly reports. Research findings come from the US Census Bureau Center for Economic Studies working paper series and the Stanford Digital Economy Lab.
Sector mapping. Challenger’s taxonomy and BLS NAICS supersectors do not align one to one. Mappings used here:
- Technology → telecommunications + computing infrastructure/data processing/web hosting + web search portals and other information services + computer systems design and related services. Publishing, broadcasting and motion picture are excluded from Technology and assigned to Media, so the two do not overlap.
- Media → publishing industries + broadcasting and content providers + motion picture and sound recording industries.
- Transportation & warehousing → Challenger’s Transportation and Warehousing categories combined, mapped to the BLS transportation and warehousing supersector, which combines them.
- Financial → financial activities. Health care → health care and social assistance. Professional & business services → professional and business services. Government → government. Retail → retail trade.
Challenger’s Health Care/Products category includes health-products manufacturers, which BLS classifies under manufacturing; this inflates health care’s cut count slightly relative to its BLS denominator, making the sector’s already-low density a conservative reading. Challenger’s Pharmaceutical category is excluded because BLS Table B-1 does not break pharmaceuticals out from chemical manufacturing at a comparable level.
AILP construction. Components, weights and normalisation are disclosed in full in the AILP section above. Min-max normalisation is applied within the eight-sector set, so component scores are relative positions within this set rather than absolute measures. Scores were computed twice, independently, from separately entered source literals; both passes agreed to one decimal place. The derived average weekly earnings figure of $1,290.37 reproduces the BLS published value exactly, confirming the earnings inputs.
Methodology versioning. Component definitions, weights and the sector set are fixed as of this baseline reading. Any future change to components or weights will be stated explicitly, re-baselined transparently, and historical readings will not be silently revised.
Scope of interpretation. Announced job cuts are employer statements of intent, not confirmed separations, and the AI attribution reflects what employers chose to say rather than independently verified causation. Cut density measures announcements against a payroll base and does not capture jobs never posted — which, per the Census and Stanford findings, is where much of the effect appears to sit. AILP is a pressure reading, not a forecast, and not a causal estimate of AI’s effect on employment. The Census adoption tier is an ordinal assignment made by Axis Intelligence Research from the published sector findings, not a continuous measure supplied by the Census Bureau.
About This Dataset
ai-job-losses-by-industry-2026.csv contains every figure cited on this page as a discrete row with full provenance: value, unit, as-of date, geography, segment, source organisation, source document, URL, retrieval date, primary-source flag, Axis-calculated flag, method note and licence.
Licence: CC BY 4.0. Attribution required.
Citation line: Axis Intelligence Research, AI Job Losses by Industry, 2026.
Cite This Research
APA Axis Intelligence Research, & Park, D. (2026). AI job losses by industry: 2026 statistics, sector rankings and displacement data. Axis Intelligence. https://axis-intelligence.com/ai-job-losses-by-industry/
MLA Axis Intelligence Research, and David Park. “AI Job Losses by Industry: 2026 Statistics, Sector Rankings & Displacement Data.” Axis Intelligence, 19 Aug. 2026, axis-intelligence.com/ai-job-losses-by-industry/.
Chicago Axis Intelligence Research and David Park. “AI Job Losses by Industry: 2026 Statistics, Sector Rankings & Displacement Data.” Axis Intelligence. August 19, 2026. https://axis-intelligence.com/ai-job-losses-by-industry/.
Frequently Asked Questions
Does an AI-attributed layoff announcement mean AI actually eliminated the role?
No. Challenger records the reason the employer states, which is a communications decision as much as an operational one. The firm maintains a separate “Technological Update (possibly AI)” category — 20,219 cuts in 2025 — precisely because attribution is often ambiguous. Treat the series as a measure of what employers are willing to say, tracked consistently over time. That consistency is what makes it useful; it is not a causal measurement.
Why does the AILP rank media second when it announced only 4,428 cuts?
Because cut announcements are 40% of the score and payroll trajectory is 35%. Media announced few cuts but recorded the steepest year-over-year payroll contraction in the set at −2.80%, and sits in the top adoption tier. A sector can shed workers steadily through attrition and non-replacement without ever issuing a layoff announcement, and media is the clearest example in this dataset.
If professional services adopts AI heavily, why is its employment growing?
Because adoption and displacement are different variables. Census BTOS data shows 66% of AI-using firms deploy it solely to augment tasks, and only 2% of firms reported AI-related employment decreases. Professional and business services added 115,000 jobs in the year to July 2026. Heavy adoption at the top of a sector is compatible with net hiring across it.
Is the 19% young-worker employment gap the same thing as a 19% job loss?
No, and the distinction matters. It is a relative measure: employment for 22-to-25-year-olds in highly AI-exposed occupations versus similarly aged workers in less-exposed occupations. In levels, Stanford reports employment for that age group in the two most-exposed quintiles fell about 11% between November 2022 and June 2026, while the three least-exposed quintiles grew about 10%. The gap is the distance between those two paths.
Why do layoff trackers and federal hiring data tell different stories?
They measure opposite ends of the same process. Trackers count separations, which generate announcements. Census and Stanford data indicate the dominant mechanism is reduced hiring of early-career workers — and a requisition that is never opened produces no announcement, no WARN filing and no tracker row. A sector can be materially affected and register almost nothing in a layoff-based measure.
Which sectors are adding jobs while AI adoption rises?
Health care and social assistance added 552,100 jobs in the year to July 2026, the largest absolute gain BLS recorded. Professional and business services added 115,000. Retail trade was essentially flat at +10,700. Challenger separately reported 107,500 announced hiring plans in 2026 through July, up 25% year over year, concentrated in aerospace and defence, technology and automotive.
Does AILP predict which industries will cut jobs next?
No. AILP is a dated reading of current pressure built from announced cuts, realised payroll change and measured adoption — all backward-looking inputs. It is designed to be recomputed as new data arrives, so that direction of travel becomes visible over successive readings. Anyone using it as a forecast is using it wrong.
What would change the rankings most in the next reading?
The BLS annual benchmark revision, which re-anchors establishment estimates to unemployment-insurance tax records. The preliminary estimate was scheduled for August 28, 2026, with the final revision landing alongside the January 2027 Employment Situation release in February 2027. Because C2 depends on year-over-year payroll change, a material benchmark revision would move trajectory scores across every sector simultaneously.