Edited by SmartRolePath Editorial. Sources cited: U.S. Bureau of Labor Statistics, arXiv, Stanford Digital Economy Lab and 1 more. Last updated November 15, 2026.
What the Bureau of Labor Statistics, Microsoft, Stanford and Yale measured, which kinds of work show the least AI exposure, and a five question test for your own role.
Short answer: No career is fully safe from AI, and the official sources say so. The evidence does show a pattern. Work that involves physically working with people, operating machinery or manual labor has the lowest AI applicability in Microsoft's data, and healthcare is projected to supply almost one third of all new U.S. jobs from 2025 to 2035. The clearest early harm is to young workers entering AI-exposed fields: Stanford found their employment 19 percent below trend, with no comparable gap for experienced workers. Economy-wide, neither Stanford nor Yale found widespread job loss yet.
Disclosure: SmartRolePath publishes this article and offers an AI powered career analysis. Every figure below comes from the sources linked at the end, not from our tools, and we list what each source cannot tell you.
Three different questions hide inside the phrase, and the sources answer different ones. Exposure asks how much of an occupation's tasks AI can assist with or complete. Displacement asks whether workers are actually losing jobs. Demand asks whether employers will keep hiring. A job can be highly exposed and still grow, or barely exposed and still shrink for other reasons.
The Bureau of Labor Statistics (BLS) is explicit about this. Its new AI exposure categories, published with the 2025 to 2035 projections on August 27, 2026, rate each occupation Low, Moderate, High or Very high. BLS says a High or Very high rating does not necessarily mean employment will decline, and that a Low rating does not mean the occupation is unaffected or safe from technological change. It also says the categories are relative to other occupations and do not separate automation, where AI replaces a task, from augmentation, where AI helps a person do it.
So this article uses "lower exposure" and "stronger demand" instead of "safe." That is a weaker promise, and it is the one the data supports.
Each source measures something different, so none of them settles the question alone. Figures were checked on November 15, 2026.
| Source | What it measures | What it found | Where it stops |
|---|---|---|---|
| BLS Employment Projections, 2025 to 2035 (August 27, 2026) | Projected job growth and decline by occupation and industry | 5.9 million jobs added, growth of 3.5 percent. Healthcare leads. Office and administrative support shrinks. BLS says uncertainty about AI's impact remains very high. | A projection, not an observation. BLS notes AI could support demand for some occupations and dampen it for others. |
| BLS AI exposure categories (August 27, 2026) | Each occupation's exposure to AI, in four categories, from five published measures | Occupations are rated Low, Moderate, High or Very high. Low means requirements do not match current AI model capabilities well. | Not a forecast of jobs or wages. Low does not mean safe. BLS says it does not measure AI related employment impacts. |
| Microsoft, Working with AI (Tomlinson and others) | How well AI chatbot use matches an occupation's work activities, from 200,000 Bing Copilot conversations | Highest scores for information workers. Lowest for work involving physical work with people, machinery and manual labor. | The authors warn that high applicability does not mean jobs will be automated or lost. One slice of the AI market. |
| Stanford, Canaries in the Coal Mine (August 2026) | Employment changes in ADP payroll data covering millions of workers through June 2026 | No widespread displacement. Workers aged 22 to 25 in AI-exposed jobs are 19 percent below where they would be, mainly through reduced hiring. | Descriptive, not causal. The pattern weakens when controlling for education and partly predates generative AI. |
| Yale Budget Lab AI tracker (updated September 15, 2026) | Occupational churn, AI usage and exposed versus unexposed workers, using Current Population Survey data through August 2026 | No clear evidence of labor market disruption associated with AI. The occupational mix is not yet changing in ways that clearly align with AI. | The Lab says findings may change as new data arrive. |
Read together: the economy as a whole does not yet show AI-driven job loss, early signs appear at the entry level of some exposed fields, and the official projections expect AI to help some occupations and hurt others.
Microsoft's researchers scored how often and how successfully people use an AI assistant for each occupation's work activities. The paper's own summary of the pattern is that the top groups are mostly information workers, while the lowest scoring groups include occupations that require physically working with people, operating machinery and other manual labor. The table shows the highest and lowest scoring minor occupation groups.
| Occupation group | AI applicability score | Position |
|---|---|---|
| Media and communication workers | 0.38 | Highest |
| Sales representatives, services | 0.35 | Highest |
| Information and record clerks | 0.33 | Highest |
| Mathematical science occupations | 0.32 | Highest |
| Tour and travel guides | 0.32 | Highest |
| Extraction workers | 0.05 | Lowest |
| Textile, apparel and furnishings workers | 0.05 | Lowest |
| Home health aides and nursing assistants | 0.04 | Lowest |
| Grounds maintenance workers | 0.04 | Lowest |
| Plant and system operators | 0.04 | Lowest |
| Forest and conservation workers | 0.03 | Lowest |
Three cautions matter before you act on this table. First, the scores come from how people used one product, Bing Copilot, and Microsoft's authors say it would be a mistake to conclude that high scoring occupations will be automated, because the downstream effects of new technology are hard to predict. Second, a low score says nothing about pay, hiring or working conditions. Third, we read these scores from version 6 of the paper, and a newer version exists, so exact values may have changed.
Projected demand is a separate signal from exposure. The BLS release covers 2025 to 2035 and projects total employment to rise from 170.3 million to 176.2 million, growth of 3.5 percent, slower than the 10.9 percent recorded over the 2015 to 2025 decade. Healthcare support and healthcare practitioners and technical occupations together are expected to account for almost one third of all new jobs.
| Occupation or group | Projected change | Context from the release |
|---|---|---|
| Nurse practitioners | Growth of 41.0 percent | Fastest growth of all detailed occupations |
| Data scientists | Growth of 34.6 percent | Among the 10 fastest growing detailed occupations |
| Solar photovoltaic installers and wind turbine service technicians | Growth of 36.5 and 29.5 percent | Among the five fastest growing, but together fewer than 15,000 new jobs |
| Healthcare support (group) | Growth of 13.3 percent | Fastest of the 22 major occupational groups |
| Healthcare practitioners and technical (group) | Growth of 8.0 percent | Second fastest of the 22 major groups |
| Sales and related (group) | Decline of 1.4 percent | BLS names incorporation of AI tools into the sales process as a factor |
| Office and administrative support (group) | Decline of 4.0 percent, or 752,100 jobs | Fastest decline and most jobs lost of any major group. BLS says automation tools, including AI powered ones, are likely to reduce demand |
| Arts, design, entertainment, sports and media (group) | No figure in the release text | BLS says generative AI software may limit demand for some jobs |
BLS summarizes the AI effect this way: growing AI adoption is expected to support demand for some occupations, while associated productivity gains may dampen employment demand for others, and the uncertainty remains very high. These are projections, and BLS says the 2026 to 2036 set will be released in 2027.
The best current evidence on actual job effects comes from payroll records. Stanford's economists used ADP data on millions of U.S. workers through June 2026 and report six facts. The two that matter most for a career changer:
The authors call these early, descriptive indicators rather than causal estimates, and note the pattern weakens when controlling for education and partly predates generative AI. The Yale Budget Lab, using a different data set, found no clear AI related footprint in its September 15, 2026 update. The two findings are compatible: a concentrated entry level effect can exist inside an economy that has not changed much overall. For how this affects someone breaking into a new field, see our guide on AI and entry level jobs for career changers.
No published list can score your exact job, so use the pattern in the evidence as a checklist. This is our own framework, built from the findings above, not a measurement. The more "yes" answers in the first four, the lower the likely exposure. The fifth tells you how to read the result.
A role that scores well on the first two and poorly on the third is the pattern the evidence supports as lower exposure with stronger demand. It is still not a guarantee.
High exposure is a reason to adapt, not a verdict. BLS itself says the ratings do not separate automation from augmentation, and Stanford's results suggest the difference is where the damage happens. Practical steps, starting with the most evidence-backed:
If your current job is changing right now, our this week checklist covers the first steps, and the longer reads on whether AI will take your job, whether your job is AI proof and future-proof careers approach the question from other angles.
SmartRolePath's free career analysis turns your experience into ranked career paths with a month by month roadmap for the one you choose, and the Career Transition Calculator compares cost, timeline and starting salary. Both are AI estimates guided by SmartRolePath's in-house scoring framework, not forecasts or guarantees, and official BLS and O*NET data is labeled separately and applies to U.S. analyses only. The methodology page explains the split. Neither tool rates a career's AI exposure, so pair them with the BLS sources above.
The two BLS pages were read directly on November 15, 2026. The Stanford and Yale figures were read from the documents' own text. Microsoft's Table 1 scores were read from version 6 of the paper (December 2025), and a newer version exists. BLS also publishes a data file of exposure categories by occupation, which we could not download, so this article does not list occupations by BLS category. Occupation level scores beyond Table 1 were not used because we could not verify them. Pages and projections change, so confirm on each site.
SmartRolePath is not affiliated with or endorsed by the U.S. Bureau of Labor Statistics, Microsoft, Stanford University, ADP or the Yale Budget Lab.
Run the free career analysis Open the transition calculator Take the 3 minute career quiz
No source ranks careers by safety, but the evidence points to a pattern. Microsoft's measure of AI applicability is lowest for work that involves physically working with people, operating machinery and manual labor, and the Bureau of Labor Statistics projects healthcare to supply almost one third of all new jobs from 2025 to 2035. The Bureau also says a Low AI exposure rating does not mean an occupation is safe.
The evidence is indirect. Microsoft's lowest scoring occupation groups include plant and system operators, grounds maintenance workers and extraction workers, which are hands-on or machine-operating jobs. The Bureau of Labor Statistics lists solar photovoltaic installers and wind turbine service technicians among the five fastest growing occupations, though together they add fewer than 15,000 jobs by 2035. Neither source says any trade is immune.
Healthcare has the strongest demand signal. The Bureau of Labor Statistics projects healthcare support to grow 13.3 percent and healthcare practitioners and technical occupations 8.0 percent from 2025 to 2035, and together they account for almost one third of new jobs. Nurse practitioners are projected to grow 41.0 percent, the fastest of any detailed occupation. Microsoft's data also gives home health aides and nursing assistants one of its lowest AI applicability scores, 0.04.
Nobody can say for certain. Microsoft's highest AI applicability scores are in media and communication, services sales, and information and record clerks. The Bureau of Labor Statistics projects office and administrative support to shrink 4.0 percent, losing 752,100 jobs, and cites AI powered automation tools. Stanford's payroll study finds the clearest early effect is fewer entry level hires in occupations where AI substitutes for human tasks.
The latest evidence does not show it. A Stanford study of ADP payroll data through June 2026 found no evidence of widespread, economy-wide job displacement, and the Yale Budget Lab's tracker, updated September 15, 2026, found no clear evidence of labor market disruption associated with AI. The Bureau of Labor Statistics projects the economy to add 5.9 million jobs from 2025 to 2035 and says uncertainty about AI's impact remains very high.
No. The Bureau of Labor Statistics states that a Low AI exposure rating does not mean an occupation is unaffected or safe from technological change, and that the ratings are relative to other occupations, not absolute. A better question is how much of your own daily work AI can already do, and whether it helps you or replaces you.
In its 2025 to 2035 projections the Bureau sorts occupations into four categories: Low, Moderate, High and Very high. It combines five published measures, three of theoretical exposure and two of observed AI use, including studies from Anthropic and Microsoft. The Bureau says the categories are not a forecast of employment growth or decline.