Which jobs are most at risk from AI?
Studies measure how much of a job's tasks AI could do, not whether the job disappears; employment still grew in nearly all occupations analysed.
Covers: Research and official statistics on occupational exposure to AI and automation, including which tasks and job categories are most affected and how exposure differs from actual job loss. Does not predict specific future unemployment numbers or give personal career advice.
Also answers: What jobs will AI replace? · Which jobs will AI take over? · How exposed are jobs to AI automation? · What jobs are most vulnerable to AI?
The short answer
Interpretation AI-prepared starting mapResearch on AI and automation exposure does not identify a single list of doomed jobs. Instead, studies measure how closely an occupation's tasks overlap with what AI or robots can do. Across this work, clerical, administrative, financial and customer-service roles are most often flagged as highly exposed, while physically and emotionally demanding, unpredictable work such as health care, skilled trades and hospitality appears comparatively resilient. Crucially, exposure is not job loss: across 2012-2019 in OECD countries, employment grew in nearly all occupations analysed and there was no clear overall relationship between AI exposure and employment growth.123
- Evidence 22
- Interpretation 3
In brief
Clerical, administrative, financial and customer service roles are most consistently flagged as highly exposed, and some knowledge and creative work is increasingly vulnerable.1
Evidence-backedPhysically and emotionally demanding, unpredictable work such as health care, skilled trades and hospitality appears comparatively resilient.1
Evidence-backedBetween 2012 and 2019, employment grew in nearly all occupations analysed and there was no clear overall link between AI exposure and employment growth.2
Evidence-backedExposure is uneven across countries: only 12 percent of workers in low-income and 15 percent in lower-middle-income countries are highly exposed, and electricity access limits effective exposure.3
Evidence-backed
At a glance
The picture in numbers
Live · updated just now
12%
12 in every 100
15%
15 in every 100
The evidence behind it
6 sources- Reviews of many studies1
- Other studies and data5
When it was published
Newest from 2026
| Source | Kind | Year |
|---|---|---|
| Labour‐saving automation: A direct measure of occupational exposure | Other studies and data | 2023 |
| Occupational Vulnerability to AI-Driven Change: The Role of Skill Composition, Task Structure, and Psychosocial Buffers. | Other studies and data | 2026 |
| Potential occupational exposure to artificial intelligence and automation among certified journeypersons in Canada | Other studies and data | 2026 |
| Artificial intelligence and employment | Other studies and data | 2021 |
| The Exposure of Workers to Artificial Intelligence in Low- and Middle-Income Countries | Other studies and data | 2025 |
| Generative AI and Job Vulnerability: A Global Review | Reviews of many studies | 2025 |
The community around it
- Contributions
- 0
- People
- 0
- Following
- 0
Nobody has added anything yet. Experience, evidence or a different view would show up here.
What it means for you
Which fits you?
Pick the situation closest to yours. Each answer says what it rests on.
If you work in clerical, administrative, financial or customer service roles
this is the category most consistently identified as highly exposed, so the relevant question is which of your tasks could be automated or augmented and which digital skills you would need to shift toward higher value-added work.12
Evidence-backedIf you work in health care, skilled trades or hospitality
these occupations are described as comparatively resilient because their work is physically and emotionally demanding and unpredictable, though trades are still being directly assessed for exposure.16
Evidence-backedIf you have strong digital skills
the OECD analysis suggests you may be better placed to use AI effectively and shift toward non-automatable, higher value-added tasks, which is one explanation for why high-computer-use occupations saw higher employment growth with greater AI exposure.2
Evidence-backedIf you have limited digital skills
the same analysis suggests you may capture fewer of AI's productivity benefits, and it found suggestive evidence of falling average hours worked in low-computer-use occupations with higher AI exposure.2
Evidence-backedIf you are assessing your own role's risk
look at task structure and job features, not just an occupation label: high-exposure clusters varied in protective job features, and some low-exposure roles with limited buffering still face occupational stress.5
Evidence-backedIf you are in a low- or middle-income country
exposure is lower on average and concentrated among women, urban and higher-educated workers, and the authors expect labour-market impacts to be more limited than in high-income countries.3
Evidence-backedThe full story · 4 chapters
01
How researchers measure exposure
AI summary:Researchers score how much of an occupation's tasks overlap with what AI or robots can do, using patents, task databases and occupation codes.
Evidence-backed: Exposure studies try to quantify how much of an occupation's work could in principle be done by AI or robots, rather than counting jobs already lost. Montobbio et al. build a direct measure by identifying robotic and labour-saving robotic patents, mapping their underlying CPC code definitions against O*NET task descriptions, and ranking tasks and occupations by text similarity. This yields fine-grained information on which tasks and occupations are most exposed, which the authors then relate to US wage and employment dynamics and to industry and geographic penetration rates.4
Evidence-backed: Other approaches use occupation-level scores. Welithotage and Nowrouzi-Kia quantify AI exposure with an Artificial Intelligence Occupation Exposure (AIOE) score and combine it with a Psychosocial Buffer Index capturing protective job features. Their principal component analysis produced five distinct occupational clusters that differ in both AI exposure and buffering capacity, with AI exposure aligning with cognitive intensity while buffering capacity dispersed independently across the skill space. Demombynes et al. use highly detailed 4-digit occupation codes, which they argue gives a more reliable mapping of AI exposure to occupation.53
Evidence-backed: Statistics Canada applies the exposure lens to a specific group: certified journeypersons in the skilled trades, where work is task-intensive and specialised. That framing matters because trades are often assumed to be insulated, yet the analysis treats them as a population whose exposure deserves direct measurement.6
02
Which occupations look most and least exposed
AI summary:Clerical, administrative, financial and customer service roles are most often flagged, while health care, skilled trades and hospitality look more resilient.
Evidence-backed: A global review of 52 studies published between 2013 and 2025 concludes that clerical, administrative, financial and customer service jobs are currently identified as those at highest risk globally. The same review finds that knowledge-based and creative jobs, previously considered safe, are increasingly vulnerable, while occupations that are physically and emotionally demanding and unpredictable, such as health care, skilled trades and hospitality, remain comparatively resilient. It also notes regional variation in risk and differences in how employers respond.1
Evidence-backed: The cluster analysis by Welithotage and Nowrouzi-Kia adds nuance to a simple high/low ranking. High-exposure clusters varied in their psychosocial buffering, suggesting structural job features can mitigate automation-related vulnerability, while some low-exposure roles with limited buffering may still face occupational stress. The authors describe AI-related occupational risk as multi-dimensional, tied to job structure rather than exposure alone.5
Evidence-backed: Geography changes the picture. In low- and middle-income countries, AI exposure is higher for women, urban workers and those with higher education, and exposure falls as country income level falls: only 12 percent of workers in low-income countries and 15 percent in lower-middle-income countries are highly exposed. Lack of access to electricity further limits effective exposure in low-income countries, leading the authors to conclude that labour-market impacts of AI will be more limited there than in high-income countries.3
03
Exposure is not the same as job loss
AI summary:Between 2012 and 2019 employment grew in nearly all occupations studied, with no clear overall link between AI exposure and employment growth.
Evidence-backed: The OECD study of 2012-2019 found employment grew in nearly all occupations analysed and no clear overall relationship between AI exposure and employment growth. Where computer use was high, greater AI exposure was linked to higher employment growth. The authors also report suggestive evidence of a negative relationship between AI exposure and growth in average hours worked in occupations where computer use is low.2
Evidence-backed: One proposed mechanism is that partial automation by AI raises productivity directly and shifts the task composition of occupations toward higher value-added tasks, offsetting direct displacement for workers with good digital skills who can use AI effectively. The authors suggest the opposite may hold for workers with poor digital skills, who may not interact efficiently with AI and therefore capture fewer of its benefits. They stress that further research is needed to identify the exact mechanisms.2
Evidence-backed: The World Bank analysis makes the same distinction explicitly: greater exposure indicates larger potential for future change in an occupation but does not equate to job loss, since it could result in augmentation of worker productivity, automation of some tasks, or both.3
04
What may moderate risk
AI summary:Digital skills and protective job features may shape whether exposure brings displacement or productivity gains, though this is not confirmed.
Evidence-backed: Digital skills appear to shape whether exposure translates into displacement or into productivity gains, according to the OECD analysis, though the authors describe this as a possible explanation rather than a confirmed mechanism.2
Evidence-backed: Job structure may matter independently of exposure. The Psychosocial Buffer Index captures protective job features relevant to occupational health, and the finding that buffering disperses independently of exposure suggests two occupations with similar AI exposure could differ in how much strain workers experience. The authors frame this as actionable for targeted interventions including retraining, workflow redesign and collaborative AI integration.5
Evidence-backed: The global review reaches a policy conclusion rather than an individual one: it recommends concerted upskilling, AI governance and inclusive transition strategies to prevent labour markets from becoming more unequal.1
Ask this Sylo
Still wondering about something?
Answers come only from this page's reviewed material, with citations, and say plainly when the page doesn't cover it yet.
More on AI
Everything on AI ›
Photo: Arlington ResearchWill AI take my job?
Will artificial intelligence take my job, and which jobs are most at risk?
Photo: imgixWhy is Amazon dropping NDAs for data center deals?
Why is Amazon dropping NDAs for data center deals, and what does it mean for AI infrastructure?
Photo: Erwan HesryWhy are mathematicians reacting to OpenAI's latest math results?
Why are mathematicians reacting to OpenAI's latest math results, and what did the model actually achieve?
Photo: Laura OckelWhat is TypeSafe Jev and how does a non-text AI model work?
Did Anthropic's AI models hack organizations during safety tests?
Photo: AbsolutVisionWhy is USA Today suing OpenAI over training data?
Behind this page
Who's adding to it, where it comes from, how it changed and what would make it better. Always open to everyone.
Discussion
Sources
Numbers match the citations in the article. A working link isn't proof that a page supports a claim; check the quoted passage and date.
- 1Generative AI and Job Vulnerability: A Global ReviewRecent Research Reviews Journal (Pokhrel et al.)Published Nov 25, 2025Checked Oct 10, 2026
“Clerical, administrative, financial, and customer service jobs are currently identified as those globally at the highest risk, while knowledge-based and creative jobs that have been considered hitherto safe are increasingly vulnerable. Conversely, occupations that are physically and emotionally demanding and unpredictable, such as health care, skilled trades, and hospitality, remain comparatively resilient. This review also explores regional variation in risks from automation, approaches to the methodological assessment of risk, and the strategic responses from employers across industries. Conclusively, this study emphasizes a set of policy recommendations targeting concerted upskilling, AI governance, and inclusive transition strategies in efforts to prevent labor markets from becoming more unequal. This systematic literature review used information obtained from peer-reviewed journals, policy reports, and organizational datasets published between 2013 and 2025. Altogether, 52 studies were thematically analyzed and comparatively mapped across sectors in line with predetermined inclusion criteria targeted at AI-driven automation and workforce vulnerability across sectors.”
- 2Artificial intelligence and employmentOECD social employment and migration working papers (Georgieff & Hyee)Published Dec 11, 2021Checked Oct 10, 2026
“Over the period 2012-2019, employment grew in nearly all occupations analysed. Overall, there appears to be no clear relationship between AI exposure and employment growth. However, in occupations where computer use is high, greater exposure to AI is linked to higher employment growth. The paper also finds suggestive evidence of a negative relationship between AI exposure and growth in average hours worked among occupations where computer use is low. While further research is needed to identify the exact mechanisms driving these results, one possible explanation is that partial automation by AI increases productivity directly as well as by shifting the task composition of occupations towards higher value-added tasks. This increase in labour productivity and output counteracts the direct displacement effect of automation through AI for workers with good digital skills, who may find it easier to use AI effectively and shift to non-automatable, higher-value added tasks within their occupations. The opposite could be true for workers with poor digital skills, who may not be able to interact efficiently with AI and thus reap all potential benefits of the technology.”
- 3The Exposure of Workers to Artificial Intelligence in Low- and Middle-Income CountriesWorld Bank, Washington, DC eBooks (Demombynes et al.)Published Feb 5, 2025Checked Oct 10, 2026
“Additionally, unlike earlier papers, the paper uses highly detailed (4 digit) occupation codes, which provide a more reliable mapping of artificial intelligence exposure to occupation. Results within countries, show that artificial intelligence exposure is higher for women, urban workers, and those with higher education. Exposure decreases by country income level, with high exposure for just 12 percent of workers in low-income countries and 15 percent of workers in lower-middle-income countries. Furthermore, lack of access to electricity limits effective exposure in low-income countries. These results suggest that for developing countries, and in particular low-income countries, the labor market impacts of artificial intelligence will be more limited than in high-income countries. While greater exposure to artificial intelligence indicates larger potential for future changes in certain occupations, it does not equate to job loss, as it could result in augmentation of worker productivity, automation of some tasks, or both.”
- 4Labour‐saving automation: A direct measure of occupational exposureWorld Economy (Montobbio et al.)Published Nov 24, 2023Checked Oct 10, 2026
“This article represents one of the first attempts at building a direct measure of occupational exposure to robotic labour‐saving technologies. After identifying robotic and labour‐saving robotic patents, the underlying 4‐digit CPC (Cooperative Patent Classification) code definitions, together with O*NET (Occupational Information Network) task descriptions, are employed to detect functions and operations which are more directed to substituting the labour input and their exposure to labour‐saving automation. This measure allows us to obtain fine‐grained information on tasks and occupations according to their text similarity ranking. Occupational exposure by wage and employment dynamics in the United States is then studied, and complemented by investigating industry and geographical penetration rates.”
- 5Occupational Vulnerability to AI-Driven Change: The Role of Skill Composition, Task Structure, and Psychosocial Buffers.American journal of industrial medicine (Welithotage & Nowrouzi-Kia)Published Sep 6, 2026Checked Oct 10, 2026
“AI Exposure was quantified using the Artificial Intelligence Occupation Exposure (AIOE) score, and the Psychosocial Buffer Index (PBI) captured protective job features relevant to occupational health. Principal Component Analysis (PCA) was used to examine the relationships among AI Exposure, PBI, and occupational skill composition.ResultsFive distinct occupational clusters emerged, systematically differentiating themselves in both AI exposure and buffering capacity. High AI exposure clusters varied in PBI, suggesting that structural job features can mitigate automation-related vulnerability, while low-exposure roles with limited buffering may still face occupational stress. PCA results revealed that AI exposure aligns with cognitive intensity, whereas PBI shows independent dispersion across the skill space.ConclusionsThese findings highlight that AI-related occupational risk is multi-dimensional and indicates the relation between automation potential and the job structure. This framework provides actionable insights for targeted interventions, including retraining, workflow redesign, and collaborative AI integration to support worker well-being in an AI-driven labor market.”
- 6Potential occupational exposure to artificial intelligence and automation among certified journeypersons in CanadaStatistics Canada Dissemination (Leanage)Published Jan 27, 2026Checked Oct 10, 2026
“Artificial intelligence (AI) and automation hold the potential to transform the nature of work, raising concerns about how different occupations may be affected. The risks associated with technological advancements are particularly relevant for the skilled trades, where work is task-intensive and specialized. This article examines potential exposure to AI- and automation-related job transformation among certified journeyperson occupations.”
How it changed
Published 1 time since Oct 10, 2026.
- Version 2Oct 10, 2026Live now
AI-prepared Starting Map from live research.
- First published version.
Help improve it
The brief is open about what's uncertain. These are the specific gaps that new material would fill.
Open questions
Does the pre-2020 pattern of no clear exposure-employment relationship still hold for generative AI, which the OECD data predates?
No answers yet
Which specific tasks within high-exposure occupations are actually automated versus augmented, and how does that split vary by employer?
No answers yet
How exposed are certified journeyperson trades in practice, given their task-intensive and specialised work?
No answers yet
Do psychosocial buffers such as job control and support measurably reduce harm in high-exposure roles, or only correlate with it?
No answers yet
Around this topic
Sylos connect: narrower topics report up to broader ones, so what's learned in one place shows up where it matters.