There is no defensible current list of occupations that artificial intelligence will certainly eliminate. The best global evidence measures which tasks AI could affect, not which job titles will disappear. The International Labour Organization’s 2025 assessment points to the greatest potential exposure in clerical work, with rising exposure in some highly digitized media, software and finance occupations. Its central finding is that job transformation is more likely than wholesale replacement because most occupations still include tasks requiring human input.
What “kill a job” can mean
Predictions often use one phrase for several different outcomes:
- Task automation: software performs a part of the work that a person previously did.
- Job redesign: the role remains, but its routine tasks shrink and review, judgment or relationship work grows.
- Fewer workers: an employer produces the same output with a smaller team after adopting AI.
- Occupational disappearance: demand for an entire occupation falls to negligible levels.
The ILO and OECD evidence cited here primarily addresses potential task exposure and changing skill demand. It does not establish a timetable for occupational disappearance or predict a guaranteed number of layoffs.
What the ILO’s 2025 global index actually says
The International Labour Organization’s Generative AI and jobs: A 2025 update and its related global index assess how much generative AI could affect the tasks within occupations. The headline figures are exposure measures:
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problems#1 Best Overall
| ILO 2025 figure | What it means |
|---|---|
| One in four workers worldwide | Workers are in occupations with some degree of potential generative-AI exposure; this is not one in four jobs forecast to disappear. |
| 3.3% of global employment | Employment falls in the index’s highest exposure category. The proportion varies by gender and national income level. |
| Mean automation score: 0.29 in 2025, compared with 0.30 in 2023 | A score in the ILO assessment, not a percentage of jobs lost. |
| Standard deviation: 0.14 in 2025, compared with 0.30 in 2023 | A measure reported with the ILO scores; it does not represent displacement. |
Clerical occupations remain the most exposed group in the index. Some strongly digitized occupations in media, software and finance also show increased exposure as models improve at handling voice, images, video and other specialized formats. These are occupational-group patterns, not guarantees for every worker in those fields.
Why a job title is a poor automation forecast
A job is a bundle of tasks. An accountant, editor, claims processor or software developer may spend part of the day producing text or manipulating structured information, but also checking facts, resolving exceptions, obtaining approval, explaining decisions, handling confidential situations or taking responsibility for an outcome. A model may perform the first part without being able to assume the rest.
The ILO index therefore uses four exposure gradients based on two characteristics: the average exposure of an occupation’s tasks and how much exposure varies from task to task. A high and consistent score means many tasks could be affected. A job with a similar average but wide variation may have a few highly exposed tasks alongside work that remains difficult to automate. Treating both cases as an identical “risk ranking” obscures the practical difference.
Rank #2
Questions that reveal the real exposure
- How much of the role is digital information processing that current AI can perform?
- Are exposed activities concentrated in a few routine tasks, or spread across most of the work?
- Which duties require contextual judgment, accountability, social interaction, physical presence or access to local knowledge?
- Would an AI output be allowed to stand alone, or must a qualified person verify and approve it?
Which occupational groups show the strongest signals?
Clerical occupations
Clerical work has the highest exposure in the ILO’s 2025 GenAI index. Many clerical tasks involve standardized documents, data entry, scheduling, classification, record retrieval or routine correspondence—activities that can be represented digitally and processed at scale. That finding identifies where employers may redesign workflows first; it does not show that every clerical occupation, employer or country will eliminate its staff.
Free tools Windows power users keep installed
One-click scans. No signup required.
Media, software and finance
The ILO reports increasing exposure in some highly digitized media-, software- and finance-related occupations. Generative systems can now work across text, code, images, audio and video, expanding the set of tasks they may assist with. Human review, domain responsibility, security, originality, client communication and compliance can still remain integral to the role.
Highly skilled professional work
OECD analysis of OECD countries places IT professionals, business professionals, managers, chief executives, and science and engineering professionals among occupations highly exposed to AI capabilities. This is an important warning against assuming that exposure is confined to low-paid or entry-level work. It also does not mean these professions are first in line to vanish: their work often includes non-routine judgment, social coordination, professional accountability and creative problem-solving.
Exposure is not the same as automation probability
“Exposure” asks whether a technology could affect tasks. “Automation risk” asks whether an employer is likely to hand those tasks to a system, under real technical, legal and organizational constraints. Adoption costs, data quality, infrastructure, workplace rules, liability, customer expectations and worker input all influence the second question.
| Measure | Question it answers | What it cannot prove by itself |
|---|---|---|
| AI capability exposure | Could current or projected AI perform or assist with parts of the work? | That an employer will deploy it, or that the occupation will disappear. |
| Observed adoption | Are organizations actually using AI for these activities? | That adoption has reduced total employment. |
| Hiring or vacancy trends | Are requested skills or advertised roles changing? | That a short-term trend will continue or reflects net job losses. |
| Employment displacement | Did the number of people employed fall after a change? | That AI alone caused the change rather than trade, demand, policy or another technology. |
Generative AI is only one source of automation. Robotics and other systems can affect physical or operational work that appears less exposed to text-and-image models, so a GenAI exposure index should not be treated as a complete forecast of all technological displacement.
What changing skill demand tells us
In its 2024 analysis of the labour market, the OECD reported that the share of vacancies in highly AI-exposed occupations asking for at least one emotional, cognitive or digital skill rose by 8 percentage points. The same work found establishment-level evidence that demand for these skills was beginning to fall. The vacancy result is therefore evidence of changing requirements at that point in time, not a guarantee that demand will keep rising.
For a worker, the practical implication is to watch the task mix and the skills attached to it: checking and documenting AI output, communicating decisions, managing exceptions, protecting data, working with customers and applying domain judgment may become more valuable even when the job title remains unchanged.
How to assess your own role without a misleading ranking
- Inventory the tasks. Write down recurring activities for a typical week, including preparation, communication, review and follow-up—not just the visible deliverable.
- Mark digital, repeatable work. Identify tasks based mainly on text, structured data, code, images, audio or video that can be supplied to software.
- Separate assistance from substitution. For each task, ask whether AI can draft or retrieve information, or whether it could complete the task without a person checking accuracy, legality, safety and context.
- Locate accountability. Note decisions for which a named person must explain the result, meet a professional rule, protect confidential information or handle an exception.
- Track workplace evidence. Look for changed workflows, approved tools, revised job descriptions and new review requirements in your organization rather than relying on a global occupational label.
- Build adjacent capability. Prioritize skills that let you supervise, verify, integrate or apply AI within your domain. No general evidence guarantees that retraining will protect any particular person’s job.
How reliable are the global estimates?
The ILO’s 2025 method combines task-level information, worker input, expert validation and AI-assisted predictions. It draws on 29,753 tasks in the Polish occupational classification and 52,558 data points on perceived automation potential for 2,861 tasks, then extends task predictions into ISCO-08 occupations. The accompanying assessment covers 436 detailed occupations and applies exposure estimates to labour-force survey data from more than 140 countries.
Those design details make the index broader than a list of anecdotes, but they do not turn modeled potential into observed job loss. Results can differ with national infrastructure, skills, employer choices, occupational classification and the pace of adoption. Country-level employment conclusions require local occupational and labour-market data.
Best Value
What the evidence cannot currently provide
- A reliable list of occupations that will certainly disappear.
- A date by which a named occupation will be eliminated.
- A country-by-country forecast of actual net job losses attributable to generative AI.
The ILO explicitly cautions that the future cannot be predicted while the technology is evolving. Its index is best used to identify where task redesign deserves attention, not to label workers as doomed.
What a responsible transition looks like
Employers and policymakers can use exposure information to examine workflows, consult affected workers, set review and accountability rules, and provide training before deployment changes jobs. The ILO’s May 2025 coverage emphasizes social dialogue. Pawel Gmyrek, the study’s lead author, said: “We went beyond theory to build a tool grounded in real-world jobs. By combining human insight, expert review, and generative AI models, we’ve created a replicable method that helps countries assess risk and respond with precision.” ILO Senior Economist Janine Berg added: “It’s easy to get lost in the AI hype. What we need is clarity and context. This tool helps countries across the world assess potential exposure and prepare their labour markets for a fairer digital future.”
Those statements describe an approach to preparation, not a promise that every displaced worker can be moved into an equivalent role. The most defensible near-term expectation is a changing mix of tasks, skills and staffing decisions, with outcomes varying by occupation, employer and country.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
