Data study · 26 July 2026 · 6 min read

The 25 jobs most exposed to AI, and the 25 least

Both ends of the AI Job Risk Index, ranked from 423 occupations scored against published research. The top and bottom of the list are not what most coverage suggests.

We scored 423 occupations for exposure to current AI, aggregating estimates from 29 peer reviewed papers and industry reports. Every occupation here is backed by at least two independent sources, because a single study is not a consensus.

Here are both ends of that list. The interesting part is not the top. It is how quickly the scores fall away, and what the two ends have in common.

The 25 most exposed occupations

Nine of the top 25 are administration roles. That is the clearest single pattern in the whole dataset. The work in these jobs is mostly reading text, moving it between systems, summarising it and answering routine questions about it, which is precisely the shape of task current models handle well and cheaply.

Notice too that the top score is 70. Not 95. We come back to that below.

The 25 least exposed occupations

Eleven of the bottom 25 are construction and trades. The rest are almost all manufacturing, physical care or outdoor work.

The dividing line is not skill, status or pay. It is whether the core of the job happens on a screen. A roofer and a bookkeeper can have similar training and similar pay. One works in an unpredictable physical environment where every job site differs, and the other works in a document.

Healthcare sits at both ends, and that is the useful lesson

The single most instructive thing in this dataset is that healthcare appears near the top and near the bottom.

Medical Records Specialists score 67. Medical Transcriptionists score 63. Both are among the most exposed occupations we scored. Meanwhile Healthcare Support Workers score 9 and Health Technologists and Technicians score 7.

Same industry. Sixty points apart. Sector is a weak predictor of exposure. The task mix inside a role is a strong one.

That is worth holding on to if you work in a field the headlines have declared safe or doomed. The headline is about the industry. Your exposure is about your day.

By sector, the spread is narrower than you would think

Sector means across the occupations we scored:

Technology sits joint top. That is not a mistake and it is not irony for its own sake: writing, reviewing and translating code is text work, and the tools were trained on a great deal of it.

The honest caveats

Three things worth saying plainly, because they change how much weight these numbers deserve.

These are occupation averages, not verdicts on people. A score describes the typical version of a role. Two people with the same job title and very different day to day work will have very different exposure. That is the entire reason the personal assessment exists.

Evidence depth varies. Some occupations here are backed by ten or more sources. Others clear our bar with exactly two. More sources means a more stable number, and every occupation page lists the specific studies behind its score so you can judge for yourself.

Exposure is not the same as job loss. Every source in this dataset measures how much of a role's task content overlaps with what AI can do. None of them measures how many people will actually be employed in five years, which depends on economics, regulation, and how firms choose to deploy the tools.

The full list of all 423 occupations, with the sources behind each one, is in the AI Job Risk Index.

Figures come from the AI Job Risk Index and were current when this was published. Scores change as new research is ingested, so the index is always the live version. See how scoring works. Informational guidance based on published research, not professional career or financial advice.