AI at Work Observatory

Macroeconomic metrics.

Who uses AI at work, in which countries, and what do the statistics show about firms, jobs and prevention?

The map compares three international sources, which do not measure the same thing. Other measures follow, each with its population, its date and its source.

Three sources, three measures.

To our knowledge, no source measures AI use at work in every country. Three come close. Choose one: the map, the ranking and the definition change with it.

See the sources and their limits

The three sources cannot be added up or averaged: they differ in who is counted, in what counts as AI and in their dates. For France, they give 44%, 24.6% and 49.6%.

Worker survey

Trust, attitudes and use of artificial intelligence: A global study 2025, University of Melbourne and KPMG. 48,340 people surveyed online in 47 countries, from November 2024 to mid-January 2025, 32,352 of them in work.

An online panel only reaches people who are online: India is at 91% here and at 18.5% in the usage estimates.

The value for France is read from the report’s ranked chart, between Belgium and Japan, both at 44%. Its country profile prints 59%, which appears to include any use, however rare.

Global report · Country profiles

Official statistics

Individuals – use of generative AI tools (dataset isoc_ai_iaiu), Eurostat, 2025 survey, updated on 5 June 2026. National statistical offices ask people aged 16 to 74 about their use of generative AI in the previous three months.

The map shows use for the job among people in work: employees, the self-employed and family workers. Only generative AI counts, so the measure is narrower than the survey’s. Counting everyone aged 16 to 74, the European Union figure falls from 21.8% to 15.4%.

Eurostat database

Usage estimates

Global AI Diffusion: Q2 2026 Trends and Insights, Microsoft AI Economy Institute, September 2026. Estimated from Microsoft’s aggregated and anonymised usage data, adjusted for market share, internet access and population.

It is the only one of the three sources that covers most of the world, and the only one that is not about work: it counts every use by people aged 15 to 64, in work or not.

36 economies share exactly the same series as their neighbours, for example twelve West African countries at 10.4%, which suggests regional estimates.

About the report · Data on GitHub

One colour scale serves all three sources. Base map: Natural Earth, with no position taken on disputed borders. Data compiled on 21 November 2026.

See all the data as tables
Worker survey (University of Melbourne and KPMG, November 2024 to January 2025). Share of people in work who say they use AI at work at least every few months. * Read from the report’s ranked chart.
RankCountryShare (%)
1India91
2China88
3Nigeria85
4Egypt84
4United Arab Emirates84
6Saudi Arabia82
7Costa Rica76
8Brazil75
9South Africa74
10Türkiye69
11Norway67
12Colombia65
12Switzerland65
14Mexico64
14Singapore64
16Argentina62
17Chile58
18Lithuania54
18Poland54
18South Korea54
21Denmark53
21Estonia53
21Latvia53
24Romania52
24United States52
26Slovenia50
26Spain50
28Australia49
28Austria49
28Italy49
28Portugal49
32United Kingdom48
33Finland47
33Ireland47
35Israel46
36Belgium44
36France44*
36Japan44
36Sweden44
40Canada43
40Greece43
42Germany42
42Netherlands42
44New Zealand41
45Hungary39
46Slovakia35
47Czechia33
All 47 countries58
Official statistics (Eurostat, 2025 survey). Share who used generative AI for their job in the previous three months.
RankCountryPeople in work (%)Everyone aged 16-74 (%)
1Norway43.835.4
2Switzerland41.034.4
3Kosovo40.924.3
4Netherlands37.126.6
5Malta37.029.5
6Finland36.925.1
7Denmark35.027.2
8Luxembourg34.824.9
9Estonia33.125.1
10Belgium32.720.8
11Cyprus30.822.7
12Sweden29.921.0
13Austria28.821.2
14Ireland28.622.4
15Greece28.116.1
16Portugal27.619.9
17Lithuania27.118.8
18Spain26.117.9
19France24.618.4
20Slovenia22.915.2
21Germany20.815.8
22Bosnia and Herzegovina20.18.6
23Latvia19.914.7
24Czechia19.613.6
25Hungary18.314.9
26Albania18.310.2
27Slovakia16.011.0
28North Macedonia15.48.6
29Bulgaria14.18.4
30Croatia14.09.0
31Poland13.38.4
32Italy12.58.0
33Türkiye10.55.8
34Romania8.65.2
35Serbia7.55.8
European Union21.815.4
Usage estimates (Microsoft AI Economy Institute). Share of people aged 15 to 64 who used a generative AI product, for any purpose. † Same series as neighbouring economies in the source.
RankEconomyJune 2026 (%)First half of 2025 (%)
1United Arab Emirates73.359.4
2Singapore64.358.6
3Ireland49.941.7
4France49.640.9
5Norway49.445.3
6Spain45.139.7
7New Zealand44.137.6
8United Kingdom43.736.4
9Netherlands43.536.3
9Qatar43.535.7
11Australia40.734.5
12South Korea40.625.9
13Belgium40.333.5
14Switzerland39.132.4
15Israel38.733.9
16Canada38.533.5
17Sweden37.331.2
18Austria35.429.1
19Taiwan33.826.4
20Hungary33.327.9
21United States33.026.3
22Denmark32.426.6
23Germany32.326.5
23Poland32.326.4
25Italy31.425.8
25Saudi Arabia31.423.7
27Czechia31.226.0
27Jordan31.225.4
29Bulgaria30.725.4
30Finland30.625.6
31Slovenia30.124.6
32Costa Rica29.725.1
33Vietnam27.921.2
34Lebanon27.724.8
35Oman27.622.6
36Portugal27.422.4
37Croatia27.221.8
38Slovakia27.122.1
39Colombia25.920.4
40Dominican Republic25.822.0
41Uruguay25.720.9
42Serbia25.319.7
43Lithuania25.121.0
44Jamaica25.022.2
45Japan24.716.7
46Panama24.220.3
47Chile24.019.6
48South Africa23.919.3
49Argentina23.417.8
50Bosnia and Herzegovina23.218.2
51Kuwait22.617.7
51Malaysia22.618.3
53Georgia21.617.3
54Greece21.517.7
55Mexico21.416.7
56Philippines20.717.1
57Ecuador20.317.0
58Brazil20.215.6
59El Salvador20.114.6
60Albania19.215.8
61Moldova19.116.6
62Türkiye18.913.4
63Azerbaijan18.714.2
64India18.514.2
65Romania18.015.3
66Mongolia17.912.6
67China17.515.4
68Guatemala17.313.7
68Peru17.313.4
70Kazakhstan16.912.7
71Gabon15.712.3
72Egypt15.512.5
72Libya15.512.7
74Namibia15.413.0
75Botswana15.012.8
76Indonesia14.811.7
77Nepal14.612.3
78Honduras14.512.4
79Senegal14.412.4
80Tunisia13.912.3
81Algeria13.811.3
82Côte d’Ivoire13.710.8
83Thailand13.49.1
84Bolivia13.310.9
84Zambia13.311.7
86Iraq13.010.3
87Iran12.89.6
88Paraguay12.610.1
89Nicaragua12.510.0
90Morocco12.010.5
91Pakistan11.99.7
92Gambia11.510.6
93Angola11.4†8.9
93Madagascar11.4†8.9
93Malawi11.4†8.9
93Mozambique11.4†8.9
97French Guiana11.3†8.3
97Guyana11.3†8.3
97Suriname11.3†8.3
97Venezuela11.3†8.3
101Benin10.4†8.7
101Burkina Faso10.4†8.7
101Ghana10.4†8.7
101Guinea10.4†8.7
101Guinea-Bissau10.4†8.7
101Liberia10.4†8.7
101Mali10.4†8.7
101Mauritania10.4†8.7
101Myanmar10.48.4
101Niger10.4†8.7
101Nigeria10.4†8.7
101Sierra Leone10.4†8.7
101Togo10.4†8.7
114Belarus10.37.6
115Kyrgyzstan10.07.6
116Lesotho9.98.8
116Russia9.97.6
118Ukraine9.59.1
119Cameroon9.1†7.0
119Central African Republic9.1†7.0
119Chad9.1†7.0
119Congo9.1†7.0
119DR Congo9.1†7.0
124Haiti8.97.1
124Kenya8.97.8
124Zimbabwe8.96.9
127Laos8.46.0
128Bangladesh8.06.5
128Burundi8.0†6.4
128Eritrea8.0†6.4
128Ethiopia8.0†6.4
128Somalia8.0†6.4
128South Sudan8.0†6.4
128Sudan8.0†6.4
128Tanzania8.0†6.4
128Uganda8.0†6.4
137Armenia7.76.2
138Papua New Guinea7.67.2
138Rwanda7.66.0
138Sri Lanka7.66.2
138Uzbekistan7.65.7
142Syria7.46.7
143Cuba6.85.7
144Afghanistan6.3†5.1
144Tajikistan6.3†5.1
144Turkmenistan6.3†5.1
147Cambodia6.14.6
World18.8—

Why track the spread of AI.

This site’s thesis: occupational health must take its place at the heart of AI Safety. Safety depends on models and on their interactions with people and organisations. The more AI spreads through work, the more this dimension matters.

A question of scale.

What happens within a team can become a societal issue when the same changes spread and persist. If they affect millions of workers, even modest individual effects could represent a major public health issue.

Two collective risks.

Widespread psychosocial hazards could cause substantial population health harm. Losses of autonomy, skills and bargaining power could signal a broader erosion of human agency, or human disempowerment. These collective risks remain hypotheses to be documented.

Only one axis measured.

The map shows the speed and scale of deployment, the first axis of the AI / prevention ratio. It says nothing about the second: risk assessment, social dialogue, the ability to act and make corrections. Other surveys come closer: see beyond adoption.

Beyond adoption.

Other sources measure what the map leaves out. Four findings, with their orders of magnitude.

Spread and uses

One EU enterprise in five uses AI, up from 13.5% a year earlier. Worldwide, one worker in four is in an occupation exposed to generative AI.

  • EU enterprises that use AI10 or more employees, 2025 · Eurostat20.0%
  • Same measure in 202413.5%
  • French enterprises that use AI2025 · Insee18%
  • Same measure in 202410%
  • Jobs exposed to generative AIWorld, 2025 · ILO1 in 4
  • Augmentation conversations on Claude.ai45% automation, November 2025 · Anthropic52%

Employment: early signals

At the scale of the economy, the monitoring available (OECD, ECB, Yale) shows no net effect of AI on employment. But hiring of young people is falling more in the most exposed occupations, in France as in the United States, although causality is not established. Not counted in these figures: the voluntary exit Microsoft offered to about 60,000 employees in October 2026.

  • Announced job cuts where employers cite AIUnited States, January to October 2026, reason given by the employer · Challenger≈ 20%
  • Employment gap for 22-25 year-olds in highly exposed occupationsUnited States, June 2026, compared with less exposed occupations · Stanford19%
  • Same gap, July 2025 data15%
  • Fall in permanent hires of 21-29 year-olds, most exposed occupationsFrance, November 2022 to March 2026 · Banque de France28%
  • Same fall, least exposed occupations8%
  • One-year fall in employment of 15-29 year-olds in IT activitiesFrance, end of 2025, excluding apprentices · Insee7.4%
  • Firms that use AI to cut labour costsShare of AI-using firms, euro area, end of 2025 · ECB15%

Perceived effects at work

Nearly one EU worker in two says digital technologies determine the pace of their work. The question covers digital technologies in general, not AI alone.

  • Digital technologies determine the pace of workEU workers, 2025 · OSH Pulse48%
  • They result in working in isolation30%
  • They increase the workload28%
  • They reduce autonomy16%

Prevention and voice at work

Consulting employees is on the rise but remains a minority practice, their participation in measures against psychosocial risks is falling, and 15% of employees had AI training in the past year.

  • Employees consulted about the effects of digital technologiesEstablishments using digital technologies; 24% in 2019 · ESENER 202435%
  • Digital technologies covered in risk assessmentsShare of establishments · ESENER 202443%
  • Employees involved in measures against psychosocial risksShare of establishments; 61% in 2019 · ESENER 202455%
  • Employees who had AI training in the past year11 EU countries, 2024 · Cedefop15%
  • Employees with no formal way to express their viewsEU, 2024 · Eurofound21%

Bars show % of the population stated, on a scale from 0 to 100; for employment, they give the size of a fall or a gap. Hollow bar: the previous measure from the same source. Each bar has its own population and year: values cannot be compared from one source to another. Figures taken on 22 November 2026. Not charted: in France, fewer than one company agreement in 1,000 mentions AI, and a study found 600 incidents affecting workers in the OECD’s monitor.

See the details and sources
Other measures of AI at work, by source
What is measuredOrder of magnitudeSource
EmployersEU and France, 202520.0% of EU enterprises with 10 or more employees use AI, up from 13.5% in 2024: from 5.2% in Romania to 42.0% in Denmark. France: 18%.Eurostat, December 2025 · Insee Première no. 2120, July 2026
Exposed jobsWorld, 2025One worker in four is in an occupation exposed to generative AI: 34% of employment in high-income countries, 11% in low-income countries. At the highest level of exposure: 4.7% of women’s employment, 2.4% of men’s.ILO, Working Paper 140, May 2025
How AI is usedWorld, a single provider, November 2025On Claude.ai, 52% of conversations are augmentation (learning, iterating, getting feedback) and 45% are automation (handing over the whole task).Anthropic Economic Index, January 2026
Layoffs attributed to AIUnited States, January to October 2026Employers cited AI for 121,401 announced job cuts, about 20% of the total; Challenger has recorded 193,226 since 2023. The reason is the one the employer gives. In October it accounted for under 3% of announcements, and Microsoft’s voluntary exit offer is not counted.Challenger, Gray & Christmas, October 2026 report
Voluntary exits at MicrosoftUnited States, Canada, India, United Kingdom, October 2026Microsoft offered a voluntary exit to about 60,000 employees, 27% of its workforce. That is the number eligible: how many will leave is not yet known. Its chief executive puts it down to a reorganisation of work around AI, not to replacing employees.Microsoft, October 2026 · Associated Press, October 2026 · TechCrunch, November 2026
Employment of young people in exposed occupationsUnited States, June 2026Employment of 22-25 year-olds in occupations highly exposed to AI stands about 19% below where it would be had it kept pace with young workers in less exposed occupations (15% a year earlier). The adjustment comes through fewer hires. A descriptive result, not a causal one.Brynjolfsson, Chandar and Chen, Stanford Digital Economy Lab, August 2026
Hiring of young peopleFrance, November 2022 to March 2026Permanent hires of 21-29 year-olds fell by about 28% in the fifth of occupations most exposed to generative AI, against 8% in the least exposed fifth. Causality is not established.Palladino et al., Banque de France, October 2026
Employment in ITFrance, end of 2023 to end of 2025Salaried employment in IT activities and information services fell by 3.0%; employment of 15-29 year-olds, excluding apprentices, fell by 7.4% in one year in IT activities. Insee considers it impossible to attribute this fall to generative AI alone.Insee, Note de conjoncture, March 2026
Overall effect on employmentOECD, euro area, United States, 2026The OECD considers the evidence of an effect of AI on young people’s employment limited so far. In the euro area, firms that use AI neither create nor destroy more jobs than others; 15% of them say they use it to cut labour costs. In the United States, Yale’s tracker detects no net effect.OECD, Employment Outlook 2026 · ECB, March 2026 · Yale Budget Lab, October 2026
Perceived effects of digital toolsEU, workers, 2025Digital technologies set the pace of work for 48% of workers, isolate 30%, increase the workload of 28% and reduce the autonomy of 16%. The question is not about AI alone.EU-OSHA, OSH Pulse 2025
Prevention in workplaces30 European countries, 202435% of establishments that use digital technologies consult employees about their impact on safety and health (24% in 2019); 43% cover them in risk assessments. Employee participation in measures against psychosocial risks is falling: 55%, from 61%.EU-OSHA, ESENER 2024
Voice at workEU, 202421% of employees have no formal way to express their views at work.Eurofound, EWCS 2024
Training11 EU countries, 202415% of employees took part in AI training in the past year; 44% do not expect their employer to train them.Cedefop, AI skills survey, 2024
Collective bargainingFrance, 2017-2024Fewer than one company agreement in 1,000 has mentioned AI since 2017, a proportion multiplied by 2.5 between 2018 and 2023.Greenan, Napolitano and Pillosio, Connaissance de l’emploi no. 200, 2024
IncidentsWorld, from press reportsA study found 600 incidents affecting workers in the OECD’s monitor, which depends on what the press reports.Ledford, Information Research, 2025 · OECD monitor

Read these figures with care.

Announcements, not departures.

The count of layoffs attributed to AI rests on the reason the employer gives. A voluntary exit offer, like Microsoft’s, counts eligible employees, not departures.

Gaps, not causes.

Studies of young people’s employment compare more and less exposed occupations. Their authors describe a gap; they do not establish that AI is its cause.

Different scopes.

Each measure has its own population, country and date. The values cannot be added up or compared from one source to another.