What ILO, IMF, OECD and US labour statistics say about AI and jobs: the most exposed occupations, the growing ones, and why no official source has a 2030 list.
Most figures quoted on this question come from exposure indexes. An index splits each occupation into tasks and rates how many of them an AI system could perform or make faster. A high score means the technology overlaps with the work. It says nothing about whether employers will adopt it, what that costs, or whether cheaper output will raise demand.
When the US Bureau of Labor Statistics (BLS) gave every occupation it projects an AI exposure category in August 2026, it attached a list of limits:
An exposure category is not a forecast of employment growth or decline. "High" or "Very high" relative AI exposure does not necessarily mean employment will decline.
The ILO calls its own figures "the upper threshold of the percentage of employment that could be affected if GenAI technology were fully implemented", and they "do not account for new jobs that may be created". Eloundou and co-authors, whose task ratings feed the BLS categories, define exposure "without distinguishing between labor-augmenting or labor-displacing effects".
The International Labour Organization published its refined index in May 2025. A survey of 1,640 workers produced 52,558 ratings of 2,861 tasks, which an expert panel reviewed and a model then extended to every occupation in the international classification, ISCO-08. The results:
Of the thirteen occupations in gradient 4, eight are clerical, among them data entry clerks, typists, bookkeeping, payroll and general office clerks. The other five are contact centre salespersons, securities dealers and brokers, credit and loans officers, financial analysts, and web and multimedia developers.
The 2025 edition moved the 2023 scores both ways. Clerical scores came down after two years of practice with the tools: taking meeting notes had been scored as high as 0.9, and the authors now write that "many of these tasks still require substantial human effort." Several highly digitised occupations went up, because models have gained multimodal abilities and "some agentic capabilities, allowing them to execute multi-step tasks autonomously". Our AI agents explainer covers what that involves.
The ILO concludes that "transformation of jobs is the most likely impact of GenAI". It adds a warning: if jobs are not helped to evolve, "even partial automation of existing tasks could lead to a decline in overall job demand".
The IMF's January 2024 Staff Discussion Note reaches a bigger number: almost 40% of global employment exposed to AI, about 60% in advanced economies, 40% in emerging markets and 26% in low-income countries.
It adds a second measure, complementarity: how far a job's social, ethical and physical context shields it from displacement. Judges are the example. Text analysis makes their work highly exposed, yet "society is currently unlikely to delegate judicial rulings to unsupervised AI", so AI should make them more productive. Clerical workers are as exposed, less shielded and more at risk. "Telemarketers are a prime example," the note says. Of the exposed jobs in advanced economies, about half "may be negatively affected by AI" and the rest could gain productivity.
The IMF also measures exposure differently. It scores the overlap between ten AI applications and 52 human skills, sets the line between high and low at the median, and treats its measures as relative. The ILO scores generative AI task by task against fixed thresholds, so the two headline shares cannot be compared directly.
The BLS 2025-35 projections, released on 27 August 2026, have US employment growing by 5.9 million jobs, or 3.5%, to 176.2 million. Office and administrative support is the fastest-declining major group, down 4.0% or 752,100 jobs, a fall the BLS links to "the continued integration of automation tools, including those powered by AI, into workflows". Healthcare support grows 13.3%, nurse practitioners 41.0% and data scientists 34.6%.
A table released the same day pairs each occupation's AI exposure category with its projection. Our tally:
A projected decline reflects many forces besides AI. The fastest-declining list also includes foundry mould and core makers, where generative AI plays no obvious part. The technical note adds: "Developments in AI are proceeding rapidly, and the uncertainty about potential impacts remains very high."
At the bottom of the ILO index, with mean task scores of 0.09 on a scale of 0 to 1, sit building construction labourers, crop farm labourers, garbage and recycling collectors and bricklayers. Generalist doctors (0.29), nursing professionals (0.25), secondary teachers (0.30) and judges (0.31) also fall in the not-exposed group.
Some occupations the BLS rates "low" grow fast: construction labourers are projected to add 109,300 jobs by 2035, and physician assistants to grow 21.1%.
The BLS adds a caveat: "'Low' relative AI exposure does not necessarily mean the occupation will be unaffected by AI or is safe from future technological change." In its words, "most sources are explicitly focused on language modeling and do not include image or video generation", and the IMF framework leaves out AI combined with robotics. Robotics is covered in our piece on physical AI.
None of these sources forecasts 2030. The ILO index is "a static snapshot of the exposure of existing occupations to GenAI at the start of 2025." Eloundou and co-authors write: "We do not make predictions about the development or adoption timeline of such LLMs." The IMF calls the pace of adoption "difficult to foresee". The BLS projects to 2035 and "releases new projections annually".
The OECD's Employment Outlook 2023 found that, with AI taken into account, occupations at the highest risk of automation hold 27% of employment across the countries it studied, again with no date. Its 2026 edition reports that unemployment among young graduates has been rising relative to the wider working-age population "since well before the spread of generative AI models".
The BLS has faced such uncertainty before. In the 2010s it left its truck driver projections unadjusted for autonomous vehicles, and employment of heavy and tractor-trailer truck drivers grew from about 1.7 million in 2012 to over 2.2 million by 2023. A 2025 BLS study notes that "autonomous trucking remains a developing technology and has not yet had any meaningful employment impacts."
The ILO places much of the outcome with employers. Whether exposure ends in job losses, it writes, "will depend on the initial decision to adopt the technology, but also the extent to which individuals in these occupations are given opportunities to learn to work with these technologies and adapt to the evolving nature of their tasks." Both turn on choices made inside firms: which tasks to automate, and how much training staff in exposed roles receive. The IMF adds that, this time, "AI displacement risks extend to higher-wage earners".
The indexes rank occupations, which tells you how much of your kind of work current AI could do. If most of your tasks are routine text, data entry or call handling, you are in the most exposed group of every index. Even there, the BLS projects most very high exposure occupations to grow to 2035.
Clerical jobs. Among the occupations it rates very high for AI exposure, the BLS projects the fastest declines for word processors and typists, switchboard operators, data entry keyers, telemarketers, order clerks and payroll clerks.
None is guaranteed. The least exposed are physical, in-person jobs: construction, the building trades, farm work, cleaning and refuse collection. The ILO also rates doctors, nurses, secondary teachers and judges as not exposed. Robotics, which these indexes do not measure, is the open question for manual work.
None of these agencies publishes that number. Their figures are exposure shares: 24% of global employment for the ILO, almost 40% for the IMF, and 27% of employment at the highest risk of automation, AI included, for the OECD. The BLS projects US employment to keep growing to 2035.
Who shares in the gains is the issue at the centre of the Davos Super Intelligence Summit 2027, which asks how increasingly powerful intelligence can be turned into shared prosperity. Its fourth keynote-led panel is Driving social impact through innovation. The summit takes place on 19 January 2027, from 16:00 to 20:00 CET, at the Grandhotel Belvedere during Davos week. Davos Premier organises it, and access is by invitation only. The programme sets out all four panels.