AI Impact Analysis · 2026 · 111 Occupations
Not every role is at risk. These 111 occupations score below 40/100 on the JobForesight AI exposure index — meaning current and near-term language models can't reliably automate the core tasks involved.
What makes a job AI-proof? Three things, mostly. Physical presence in unpredictable environments — a plumber crawling under a sink, a paramedic on a stretcher, a glazier on scaffolding. Multi-step social judgement where the cost of being wrong is high — therapists reading between the lines, social workers weighing competing accounts, judges deciding what a reasonable person would do.
And regulated accountability — work where a human has to sign their name to the outcome and answer to a regulator, a court, or a coroner if something goes wrong. Current language models can draft the document; they can't be the licensed person who takes responsibility for it.
The roles below score below 40/100 on the JobForesight exposure index. Most of them combine at least two of those three properties. We've also given each one a time window — because AI-proof today doesn't mean AI-proof in 2030, and we'd rather show you the 5-year horizon honestly than pretend any job is permanently safe.
Last updated: October 2026.
Three structural properties make a job hard for current AI to replace. Physical presence in unpredictable environments — a plumber crawling under a sink, a paramedic stabilising a patient on a stretcher, an electrician threading wire through a wall cavity. Multi-step social judgement where the cost of being wrong is high and the right answer depends on reading subtle cues — therapists hearing what isn't said, social workers weighing competing accounts of the same incident.
And regulated accountability — work where a human has to answer to a court, a coroner, or a professional regulator if something goes wrong. AI can draft the document; it can't be the licensed person who takes responsibility. The roles above all combine at least two of those three properties — that's what makes them durable. See jobs AI will replace for the other side.
On our index, 111 of 334 roles score under 40/100, which we treat as safe for now. The safest fifteen run from surgeon (11/100) and early years educator (12), through barber, chief executive and judge (15 to 16), to chef, plumber, police officer and primary school teacher (18). Each carries a time window. None is permanently safe.
The five lowest scores we track: surgeon (11/100), early years educator (12), barber (15), chief executive officer (15) and judge (16). Their windows run from 3 to 5 years for surgeons and judges to 6 to 15 years for barbers. What they share: physical work in unpredictable settings, high-stakes judgement about people, or a named person who is accountable for the decision.
Three kinds, on our data. Hands-on work in unpredictable places: plumber (18/100), carpenter (20), firefighter (20), electrician (22). Judgement about people where mistakes cost a lot: early years educator (12), police officer (18), physiotherapist (19), social worker (22). Accountability for the decision: surgeon (11), judge (16), doctor (22).
Among the lowest-scoring roles, the trades have some of the longest windows we publish: barber 6 to 15 years, firefighter 5 to 15, plumber 5 to 10, carpenter 5 to 10. Robots struggle with unplanned physical work in someone else's space, which is most of a trade. Office roles that score low, such as chief executive (15) or judge (16), are safe from language models for a different reason: someone has to be accountable for the decision.
Judged by score and window together: barber (15/100, 6 to 15 years), plumber (18, 5 to 10), carpenter (20, 5 to 10), firefighter (20, 5 to 15), chief data officer (20, 5 to 8) and early years educator (12, 4 to 6). A low score says little of the work can be automated today. A long window says we expect displacement to start later. Both matter.
No. Our lowest score is 11/100, not 0. Every role carries a time window, from about a year for the most exposed of the safe roles up to 15 years, and the window shows when we expect displacement to start to be felt. 'AI-proof' in 2026 means the core tasks cannot be automated reliably today, not that the role never changes. Each role page shows when it was last updated.
Get your personalised AI Career Score — adjusted for your seniority, location, sector, and AI tool usage. Takes under 3 minutes.
Get Your Personalised Score →Free · No signup · 334 occupations