At 10:00 a.m. on Thursday, August 27, the Bureau of Labor Statistics released its ten-year employment projections for 2025–2035. Alongside them, with much less fanfare, it released something it has never published before: an official relative AI exposure category — Low, Moderate, High, or Very high — for every one of the 831 detailed occupations it projects.
For the first time, there is a federal statistical answer to the question every trade student has been asked at a family dinner since 2023.
The trades come out at the bottom of the scale. But the reason that matters is not the reason most people will give, so it is worth being precise about what BLS built.
What BLS actually measured
BLS did not run its own study. It combined five external measures and normalized them against each other.
Three are theoretical: work by Felten, Raj and Seamans on whether AI capabilities map to occupational abilities; Eloundou, Manning, Mishkin and Rock on whether large language models could cut the time a task takes by half; and Eisfeldt, Schubert, Taska and Zhang applying a similar framework with GPT-3.5 Turbo.
Two are observed usage: a measure built by Anthropic from real Claude conversations and API traffic mapped to O*NET tasks, and one built by Microsoft from Copilot data mapped to work activities.
BLS converted each source to a percentile rank so the five sit on the same 0-to-1 scale, took the median of the three theoretical ranks and the median of the two observed ranks, and clustered occupations into four groups on those two dimensions. Of 4,155 possible occupation-source combinations, 3,944 were observed and 211 were imputed, affecting 75 occupations.
Three things about this must travel with every number below. BLS states all three itself, in a limitations section on the same page:
- The categories are relative, and they land close to quartiles — 213 Low, 206 Moderate, 206 High, 206 Very high. “Low” means low compared with other occupations. In the agency’s words, “‘Low’ relative AI exposure does not necessarily mean the occupation will be unaffected by AI or is safe from future technological change.”
- It is not a forecast. “Exposure does not imply job loss, productivity gains, automation probability, or wage effects.”
- The technology snapshot is already old. “The theoretical sources conceptualize AI capabilities as those available no later than mid-2023,” and most of the inputs do not cover image or video generation at all.
Anyone who turns this dataset into “AI-proof jobs” is misreading it. What it supports is a comparison — and the comparison is stark.
The trades numbers
Every figure in this section was computed directly from the BLS data file, not from a secondary account of it.
| Occupational group | Occupations | Low | Moderate | High | Very high | Low + Moderate |
|---|---|---|---|---|---|---|
| All detailed occupations | 831 | 213 | 206 | 206 | 206 | 50.4% |
| Construction and extraction | 60 | 53 | 6 | 1 | 0 | 98.3% |
| Production | 105 | 58 | 43 | 3 | 1 | 96.2% |
| Installation, maintenance, repair | 51 | 21 | 27 | 3 | 0 | 94.1% |
| Healthcare support | 17 | 9 | 7 | 1 | 0 | 94.1% |
| Transportation and material moving | 50 | 23 | 22 | 5 | 0 | 90.0% |
| Healthcare practitioners and technical | 71 | 15 | 25 | 27 | 4 | 56.3% |
| Typical entry: bachelor’s degree | 177 | 0 | 11 | 77 | 89 | 6.2% |
Read the last two rows together. Among the 177 occupations where a bachelor’s degree is the typical entry requirement, not one is rated Low, and 89 — more than half — are rated Very high. Across all occupations that do not require a four-year degree, 72.5% are Low or Moderate. Across those that require a bachelor’s or more, 10.7% are.
The cleanest single result in the file concerns apprenticeship. BLS records a typical on-the-job training path for each occupation. All fifteen occupations whose typical training is an apprenticeship are rated Low or Moderate. None is rated High. None is rated Very high.
Twelve are Low: carpenters, sheet metal workers, structural iron and steel workers, brickmasons and blockmasons, glaziers, millwrights, elevator and escalator installers, reinforcing iron and rebar workers, boilermakers, stonemasons, musical instrument repairers, and terrazzo workers. Three are Moderate: electricians, plumbers, pipefitters and steamfitters, and mechanical insulation workers.
Where your program lands
This is the part worth screenshotting. Every one of the programs this site covers most heavily sits in the bottom two categories — with one exception, which is named rather than buried.
| Occupation | AI exposure | 2025–35 growth | Median wage |
|---|---|---|---|
| HVAC mechanics and installers | Moderate | +10.9% | $61,010 |
| Electricians | Moderate | +9.2% | $63,190 |
| Medical assistants | Moderate | +12.9% | $45,690 |
| Hairdressers and cosmetologists | Moderate | +7.7% | $35,790 |
| Plumbers and pipefitters | Moderate | +6.8% | $63,800 |
| Automotive service technicians | Moderate | +5.0% | $50,620 |
| Heavy and tractor-trailer truck drivers | Moderate | +3.8% | $58,640 |
| Dental assistants | Low | +7.4% | $48,070 |
| Phlebotomists | Low | +6.8% | $45,230 |
| Surgical technologists | Low | +5.0% | $64,650 |
| Diesel and bus/truck mechanics | Low | +3.6% | $61,770 |
| Licensed practical and vocational nurses | Low | +3.0% | $64,400 |
| Welders, cutters and brazers | Low | +2.4% | $53,750 |
| Pharmacy technicians | High | +6.4% | $45,750 |
Pharmacy technicians is the outlier, and it is a common trade-school health program. It is rated High while growing 6.4%. That combination is exactly what BLS says the categories cannot resolve: high exposure is not a decline forecast, and this occupation is projected to grow faster than the 3.5% all-occupation average.
There are others worth knowing about. Within construction and installation trades, the four occupations rated High are first-line supervisors of mechanics and installers (+4.1%), construction and building inspectors (0.0%), telecommunications equipment installers (−3.3%), and computer, ATM and office machine repairers (−3.0%). The pattern is legible: the exposed trade roles are the inspection, documentation and supervision ones — the parts of the job done at a desk. Outside those groups, computer numerically controlled tool operators are rated High and projected to decline 9.4%.
Two of the five fastest-growing occupations in the entire economy are trades: solar photovoltaic installers at +36.5% and wind turbine service technicians at +29.5%, both rated Moderate. Both are small — together they add fewer than 15,000 jobs by 2035.
The other half of the release
A rating means nothing without the comparison, and the projections release supplies it.
Office and administrative support is the fastest-declining major occupational group in the economy — down 4.0%, shedding 752,100 jobs through 2035, more than any other group. BLS is unusually direct about the cause: “the continued integration of automation tools, including those powered by AI, into workflows is likely to reduce demand for several office and administrative support occupations.”
Now cross that against the exposure file. Office clerks (−6.0%), secretaries and administrative assistants (−6.0%), customer service representatives (−5.3%) and retail salespersons (−0.3%) are all rated Very high. Sales occupations overall decline 1.4%, which BLS attributes partly to “incorporation of AI tools into the sales process.”
That is the actual story. It is not that the trades are safe. It is that the occupations BLS explicitly names as losing work to AI, and the occupations it rates most exposed to AI, are the same set — and the trades are not in it.
The part that cuts the other way
Here is the correlation nobody promoting this dataset will mention.
Sort the 831 occupations by exposure category and take the median of their median wages:
| Exposure | Median of median annual wages |
|---|---|
| Low | $49,120 |
| Moderate | $52,760 |
| High | $73,985 |
| Very high | $78,105 |
Low exposure tracks with lower pay, by roughly $29,000 at the extremes. Much of the Low category is short-term-training work — laborers, cleaners, food prep, packers — where AI exposure is low because the work is physical and low-paid, not because it is skilled.
Which is precisely why the apprenticeship result matters. Those fifteen occupations break the correlation: fourteen of the fifteen pay between $57,080 and $109,910, at Low or Moderate exposure. (The exception is musical instrument repairers and tuners, at $46,420.) Elevator and escalator installers and repairers is the highest-paid Low-exposure occupation that does not require a bachelor’s degree — $109,910, growing 5.7%, entered through an apprenticeship. Electrical power-line installers, also Low but entered through long-term on-the-job training rather than a registered apprenticeship, pays $95,320 and is the fastest-growing Low-exposure trade occupation at +10.3%, with 10,900 openings a year.
The claim the data supports is not “skip college, AI is coming for it.” It is narrower and more useful: among occupations with comparable pay, the ones entered through skilled trade training and apprenticeship are markedly less exposed than the ones entered through a four-year degree.
Jim Pauley, president and CEO of the National Fire Protection Association, put the sentiment side of this well when NFPA published its annual skilled trades survey in January, based on responses from 512 U.S. skilled-trade workers: “For years, the skilled trades industry has been called ‘automation-proof,’ but AI isn’t here to replace the craft; it’s here to remove the friction.”
What this means for students
Use the exposure category as one input among several, in this order: the openings number first, the wage second, the exposure category third. An occupation with 120,000 annual openings and Moderate exposure is a better bet than one with 400 openings and a Low rating.
Do not treat Low as a guarantee, and do not treat High as a disqualification. Pharmacy technicians is rated High and is growing faster than the economy as a whole. Nurse practitioners, the single fastest-growing occupation in the projections at +41.0%, is also rated High.
If you want the number for a specific job, the BLS data file is a public download listing all 831 occupations with wages, growth and openings in the same rows. The refreshed Occupational Outlook Handbook now carries the 2025–35 numbers across roughly 600 occupations in more than 300 profiles.
What this means for schools
Two cautions. The exposure category is going to appear in program marketing within weeks, and BLS’s limitations page is going to be left off the slide. A school that cites “Low AI exposure” without the word relative is making a claim the source does not support, and a prospective student with the spreadsheet can check it in about ninety seconds.
The second is that this data will age badly and visibly. BLS caps its theoretical inputs at mid-2023 capabilities and will publish the 2026–36 round in 2027. Anything printed now should be dated on the page.
Also this month
The rules governing your accreditor are being rewritten. On August 20 the Department of Education proposed revising 34 CFR part 602, the regulations under which the Secretary recognizes accrediting agencies — including ACCSC, ABHES and COE, which accredit most trade schools. Recognition is what makes a school’s students eligible for federal aid. The docket is ED-2025-OPE-1042 and comments close September 21, 2026.
Three ACCSC actions. Epic Flight Academy (New Smyrna Beach, Florida) voluntarily withdrew effective August 13 while Accredited; The Landing School (Arundel, Maine) withdrew August 4 while Accredited on Warning; JNA Institute of Culinary Arts (Philadelphia) closed effective June 30. A voluntary withdrawal in good standing is not a closure, and ACCSC’s rolling list does not say what prompted any of them.
The building trades signed an AI-infrastructure MOU. On August 10, North America’s Building Trades Unions, BlackRock and the AI Infrastructure Partnership signed a memorandum to expand apprenticeship utilization and give NABTU visibility into the project pipeline for workforce planning. It is explicitly nonbinding and commits to no numbers — the parties will “explore opportunities.” Worth noting mainly for the shape of it: the same AI buildout driving BLS’s fastest-growing industries is the one the building trades are now planning their workforce around.
What to watch
- Monday, September 21 — comments close on the accreditor-recognition proposed rule. The only hard deadline on the board.
- September 23–24 — the National Advisory Committee on Institutional Quality and Integrity meets in person, two days after that window shuts.
- October 19 and October 26 — the Wagner-Peyser staffing rule takes effect, then the rescission of the Executive Order 11246 contractor regulations.
- Workforce Pell, still open. In August we predicted at least one state would extend its first application window past its posted deadline. Colorado and Minnesota closed August 15; Illinois and South Dakota closed August 31. We will have the count next issue.
Sources
- U.S. Bureau of Labor Statistics — “Artificial Intelligence (AI) Exposure Categories” — August 27, 2026 — bls.gov/emp/publications/ai-exposure-categories.htm
- U.S. Bureau of Labor Statistics — “AI Exposure Categories and 2025–35 Employment Projections” (data file) — August 27, 2026 — bls.gov/emp/ind-occ-matrix/ai-exposure-categories.xlsx
- U.S. Bureau of Labor Statistics — “Employment Projections — 2025-2035,” USDL-26-1422 — August 27, 2026 — bls.gov/news.release/ecopro.nr0.htm
- Jim Pauley, National Fire Protection Association — “Skilled Workers Look to Technology Amid Workforce Shortages and Codes and Standards Rollbacks” — January 15, 2026 — nfpa.org
- North America’s Building Trades Unions — “NABTU, BlackRock, and AI Infrastructure Partnership Launch Strategic Collaboration” — August 10, 2026 — nabtu.org
- U.S. Department of Education — “Accreditation, Innovation, and Modernization: The Secretary’s Recognition of Accrediting Agencies,” Federal Register — August 20, 2026 — federalregister.gov
- Accrediting Commission of Career Schools and Colleges — “Voluntary Withdrawals / School Closures” — accessed September 1, 2026 — accsc.org
All employment, wage and exposure figures in this article were computed from the BLS data file linked above. Wages are median annual, 2025. Growth figures are projected change 2025–2035.
