Jobs at Risk From AI: A Realistic List, Not a Scary One

You read a headline saying millions of jobs will vanish within a few years, and you scan the list for your profession. If you do not find it, you relax for an hour. If you do, you lose sleep. Then you close the page with no idea what to do with what you read.

The confusion comes from the fact that most lists answer a question that does not help you. This is an attempt at a more realistic one, built on reports and studies whose original texts I went back to, listed at the end. The figures are current to October 2026, because this topic moves fast.

Why "Will my job disappear?" is the wrong question

Because a job is not one block. It is a bundle of tasks: some you repeat without thinking, some need your judgment, and some only happen when you are in front of another person. A tool does not take the whole bundle at once. It starts with whatever is easiest to turn into text, numbers and templates.

That is why the picture so far is calmer than the headlines. The World Economic Forum's Future of Jobs Report 2025 expected 170 million jobs to be created and 92 million displaced by 2030, a net gain of 78 million. But that is a forecast from employers, not a measurement of what has happened. In the same survey, 40% of employers said they expect to reduce their workforce where AI can automate tasks.

What has actually happened is something researchers can measure. A study Anthropic published in March 2026 found no systematic rise in unemployment among the workers most exposed to AI since late 2022. A Stanford paper from August 2026, using US payroll data through June 2026, found no evidence of widespread job displacement across the economy.

That is not full reassurance, though. The same paper found that employment of workers aged 22 to 25 in exposed occupations now stands about 19% below where it would be had it kept pace with their peers in less-exposed occupations. The gap shows up mainly through less new hiring, not layoffs of people already employed. The authors call it an early indicator rather than proof of cause, and it describes the US market, not ours.

So the useful question is not "Will it disappear?" It is: which parts of my work change first, and which part stays mine?

Tasks, not titles: how studies measure job exposure

Serious studies follow roughly one method. They break an occupation into its tasks and ask about each: can a tool do this or speed it up? The result is called "exposure", the overlap between what the tool can do and what the occupation requires. It is a possibility, not a decision an employer has made.

Read three well-known numbers through that lens:

• Most workers are touched a little, a few a lot. A 2023 paper estimated that about 80% of US workers could have at least 10% of their tasks affected by large language models, and about 19% could see half or more of their tasks affected.
• Exposure is not risk. In a January 2024 note, the IMF estimated that about 40% of jobs worldwide are exposed to AI: about 60% in advanced economies, 40% in emerging ones and 26% in low-income countries. But it separated jobs that are exposed and well protected, like a judge, whose rulings society does not hand to an unsupervised tool, from jobs that are exposed and less protected, like clerical work. It estimated that in advanced economies roughly half of the exposed jobs may be hurt and the other half may benefit.
• Clerical work is the most exposed. The ILO's 2025 update found that one in four workers worldwide is in an occupation with some exposure to generative AI, and only 3.3% of employment sits in the highest exposure category. It concluded that transformation of jobs is more likely than wholesale replacement.

Take one rule from this: the number that matters to you is not your profession's percentage but the share of your own time spent on exposed kinds of tasks.

The most exposed jobs in our Arab market

Data for our region is thinner than for the US, but it is not absent. An ILO report from September 2025 covering twelve Arab countries estimated that about 2.2% of jobs (roughly 1.2 million) are exposed to automation by generative AI, and about 14.6% (roughly eight million) could benefit from it as an assistant rather than a replacement. The share exposed to automation was 5.3% of women's jobs against 1.6% of men's, reflecting women's concentration in routine clerical work.

In May 2026 ESCWA and the ILO published a joint report on regional employment to 2035. Its modelling suggests automation may reduce employment in clerical work, sales, retail and routine administrative functions, and increase jobs in health, education, management and technology-intensive activities. In the WEF survey, employers in Saudi Arabia expect the share of work tasks delivered mainly by technology on its own to reach 45% by 2030, above the global average.

Put these sources together with the WEF's "fastest-declining" list and the same profile keeps recurring:

• Data entry and clerical work: data entry clerks, postal clerks, bank tellers, cashiers and ticket clerks, all on the WEF list.
• Secretarial and routine accounting work: administrative assistants and executive secretaries, accounting, bookkeeping and payroll clerks, and legal secretaries.
• Scripted sales and service: telemarketing and customer service built on repeated questions. Customer service representatives came second among the most-covered occupations in Anthropic's data, after computer programmers.
• Translation, writing and editing: Microsoft researchers put interpreters and translators at the top of the 25 occupations where AI is most applicable to the tasks, with writers, editors and proofreaders alongside. Graphic designers are also on the WEF list.

But "most exposed" does not mean "closest to disappearing." The Microsoft researchers themselves say their measure shows where work will change, not who will lose it. They cite the ATM: it automated a core task of the bank teller, yet the number of tellers rose when banks opened cheaper branches. Today the WEF puts bank tellers on its fastest-declining list. History does not promise a single outcome, so follow your own tasks, not the headlines.

The jobs furthest from the risk, and why

At the other end, Anthropic found that about 30% of workers have tasks that barely appear in its usage data for Claude, among them cooks, motorcycle mechanics and lifeguards. That reflects how rarely the tool is used in their work, not that it cannot be. The WEF expects the largest growth in absolute numbers in frontline work such as farm work, delivery, construction and food processing, and in care jobs such as nursing, social work and counselling. In our region the ILO report expected net gains in construction, transport and storage, teaching, personal services and engineering.

Why do these jobs resist? Three reasons recur in the research:

• The body: a tool can write the repair steps, but a hand has to do the repair. The IMF counts the physical context of an occupation, along with its social and ethical context and its skill level, in its "shielding" measure.
• Accountability and trust: where a person has to stand in front of a patient, client, judge or student and answer for the result, handing the decision to an unsupervised tool is hard.
• Knowledge gained by practice: Stanford's researchers suggest tools are better at replacing codified knowledge, the kind taught and written into procedures, and worse at replacing what comes from experience, mentoring and repeated situations. They observed, descriptively and not as proof, that entry-level hiring grew more slowly in occupations that lean on codified knowledge.

This protection is not permanent. By a theoretical measure from a 2023 paper that Anthropic builds on, models could in principle speed up 94% of tasks in computing and mathematics, while real Claude usage covers 33% of them. That gap may narrow as tools spread, so read the list as a map of today, not a verdict on tomorrow.

If you are a student or a parent weighing a major, a September 2026 study by the Federal Reserve Bank of Dallas on Texas graduates found that the most exposed majors, including computer science and languages, were associated with a relative decline in first-year employment rates, while nursing, education and psychology were among the least exposed. But it covers one state, counts only jobs held in Texas, and offers a statistical estimate, not proof. Instead of judging a whole major, ask: what tasks will its graduate spend the first year on?

What I have seen inside teams: what was replaced, what grew

This section comes from my experience, not from a study. I have worked with technology teams in Saudi Arabia, Canada and Jordan, and I have taken part in hiring and in bringing tools into daily work. Read it as a field note, not a statistic, and do not base a decision on it alone.

The first thing I noticed is that what gets automated is tasks, not jobs: a first draft of a document, a meeting summary, sorting incoming requests, writing routine tests, cleaning spreadsheets, a first-pass translation. Most often the routine part of a job shrank and its other part grew, rather than the job being cancelled outright.

And what grew? Every role that stands at the end of the chain: the person who reviews the output and says it is right, the person who knows the client or the system well enough to catch the error first, the person who states the requirement precisely enough for the tool to produce something useful, and the person who answers for the decision. Demand rose for judgment and accountability, and fell for the first draft of production.

This fits an explanation Stanford's researchers offer for the hiring pressure on beginners: the tools handle documented, checkable tasks well, and those are the tasks that used to justify having beginners on a team. So newcomers are now asked to show judgment earlier. The question I find separates candidates today is not "What can you produce?" but "How do you make sure what you produced is correct?"

A 15-minute exercise: audit your own week

You can now do for yourself what the studies do, on a single case: you. You need no software, only a sheet of paper, your calendar and your sent mail for the last week.

As you colour, you will notice you are applying the studies' own logic to your week, not to a general average.

Your next step if your job is on the list

If most of your hours are red, do not start by changing your profession. Start by changing your position inside the same task:

1. Move from execution to review and decision. Instead of being the person who writes the first report, be the one who checks it and stands behind it. That role is closest to the "protection" the studies describe: responsibility that is hard to hand to a tool.
2. Try the tool on your own work before someone imposes it. Pick one red task this week and measure time and errors with and without the tool. Then you are the author of the story, not its subject.
3. Build what cannot be copied. Knowledge of your clients, your organization's context, your reputation for accuracy. According to the research on tacit knowledge, these are the hardest for a tool to replace.
4. Choose a neighbouring role. Ask which role near yours stands at the end of the chain, where output is reviewed and the decision is owned.

And do not make a big decision, like resigning or switching majors, from a headline. Make it from your task map after a few weeks of trying.

That is what I can say with numbers. What your market will look like in five years I cannot guarantee, and nobody can. But your task map is in your hands, your experiment with the tool is in your hands, and the job market is not. That split between "what is mine" and "what is not mine" is the heart of the chapter "If Only": The Word That Steals Your Sleep of Resign From Yourself, and it helps when worry circles around an outcome you do not control.

And if you are a manager building a team at this moment, The Blind Manager has a whole axis on smart hiring and choosing the right talent. You have fifteen minutes this week that are enough to learn where you stand: open your calendar and begin.