hilighter Logohilighter Logo
EN
Get StartedSign In

ArticleFuture of work

AI Exposure in Viet Nam Is Not a Layoff Count

The ILO says 20.8% of employment in Viet Nam, about 11.5 million jobs, may be exposed to generative AI. That is not a layoff forecast.

Christian Kosiek•Published •21 min read

A woman in a garment workshop measures a paper pattern on green fabric, with a sewing machine, thread and bolts of cloth behind her
Illustrative image supplied with this article: a person measures a pattern in a garment workshop. It is not a named worker, an ILO respondent, a factory in the sources, or a hilighter screen. The file was not labelled as a photograph of a real person or as AI-generated. The scene does not show the 20.8% estimate. In the ILO brief, most manufacturing jobs have no generative-AI exposure, and clerical work holds most of the highest-exposure category.

Eleven and a half million is a large number. It becomes the wrong emergency if a reader treats it as layoffs. On 16 April 2026 the International Labour Organization published "Generative AI and jobs in Viet Nam: Labour market exposure and policy considerations." The brief's key points say that around 20.8% of employment in Viet Nam, equivalent to 11.5 million jobs, is potentially exposed to generative AI. Those key points also say the index measures task-level exposure inside occupations, rather than actual adoption or employment effects. A news page the same day says around 11.5 million women and men, one in five workers, are in occupations whose tasks are potentially exposed. The body of the brief says about one in five jobs: 20.8% of total employment, or 11.5 million workers. The pages use "jobs" and "workers" for the same 11.5 million figure. None of those sentences is a count of dismissals.12

On 2 November 2026 the ILO returned to the figure after policy discussions in Viet Nam from 26 to 30 October. The page says the April analysis estimates that 20.8% of employment, equivalent to 11.5 million jobs, is potentially exposed, whereas only around 1.8% of jobs fall into the highest exposure category. It says exposure does not necessarily translate into job displacement. The actual effects, it says, will depend on where and how businesses deploy AI, the tasks workers perform, and the support available to help them adapt. The page does not present a new November labour-force survey.3

What 20.8% includes, and what 1.8% is

The brief applies a global index of occupational exposure, the one in Gmyrek and colleagues, ILO Working Paper No. 140 (2025), to Viet Nam's 2024 Labour Force Survey. This article did not open that working paper. It uses the brief's account of the method. Occupations are sorted into six gradients. Gradients 1 to 3 are low to above-moderate exposure, where the brief says occupations are more likely to face task augmentation than full automation. Gradient 4 is the highest category: tasks that are highly susceptible and relatively standardized, and the greatest automation risk inside this index. Gradients 5 and 6 are minimal exposure or none. The 20.8% is gradients 1 to 4 added together. It is not a share the brief has already classed as full automation.1

Only 1.8% of jobs, around 1.0 million, fall into Gradient 4. The body of the brief says 1.8% of workers. The 16 April news page says around one million workers are in occupations where tasks are both highly susceptible and standardised, and that the risk of full automation affects less than 2% of the workforce, a smaller share than in Indonesia, the Philippines and Thailand. "Less than 2%" is that page's rounding of the 1.8% category. It is not a second survey, and it is not 11.5 million.12

What the April brief supports What it does not support
Around 20.8% of employment, 11.5 million jobs in the key points and 11.5 million workers in the body, is potentially exposed (gradients 1 to 4) A forecast of 11.5 million layoffs, or a count of jobs already cut
1.8% of jobs, around 1.0 million, are in Gradient 4, the highest category A finding that 1.8% of workers have already been dismissed
The index measures task-level susceptibility, not adoption or employment effects Proof that a task will be bought, that it will work in Vietnamese, or that a small firm can afford it
The 16 April news page says the more probable outcome is task transformation inside jobs, rather than large-scale displacement A guarantee that job quality, hours, learning or bargaining power will stay the same

A separate ILO research brief, the PDF headed February 2026, is titled "Workers' exposure to AI: What indicators tell us and what they do not." It says exposure indicators reveal technological susceptibility, not labour-market outcomes. They capture what AI could do under a static view of tasks, not whether a firm will find automation profitable, how workflows will change, or how employment, wages and demand will adjust. The brief says these measures cannot be interpreted as predictions of job displacement, productivity gains or reskilling needs. That paper reviews indicators in general. It is not a Viet Nam headcount. The Viet Nam brief's footnote dates a research brief of this title to March 2026. This article uses the PDF that was opened, which is headed February 2026.4

The Viet Nam brief also says the global index was built on detailed occupational codes from the Poland Labour Force Survey. Development, informality and the tasks inside an occupation mean some aspects may not fully reflect the local labour market. The authors present the figures as a basis for dialogue and further country work, not as a staffing plan for a 25-person firm.1

The large shares are not the same share

Women face higher potential exposure than men: 24.1% compared with 17.8%. They account for 55.5% of potentially exposed jobs, which the brief says largely reflects their occupational distribution. Women are also at 2.0% in Gradient 4, against the male overall rate of 17.8%, and they account for 52.2% of Gradient-4 jobs. The brief does not print a male Gradient-4 rate in that sentence. After individual and job characteristics, women remain more susceptible, with a predicted gap of 6.7 percentage points (26.0% compared with 19.3%). That prediction is not the raw 24.1 and 17.8. The April news page says women remain significantly more exposed after education, occupation and other characteristics, and it does not print 6.7, 26.0 or 19.3. It links the pattern to clerical, administrative and service roles with highly standardized, documentation-intensive tasks. In sales occupations with high employment, where women make up around two thirds of the workforce, the same page says generative AI could augment productivity. In routine clerical roles, it says the risk of task displacement is more acute. None of these rates is a share of women who will lose their jobs.12

A different dataset, on US occupations and gender, is already on this site. See where the AI jobs gender gap is widest without a degree. Those percentages are not 24.1 and 17.8, and they should not be averaged with them.

Clerical support workers are the concentration inside Gradient 4. 64.9% of clerical employment sits in that gradient. In absolute terms, clerical support workers are 0.8 million people, or 84.4% of all Gradient-4 jobs. The April news page says nearly two thirds of clerical support workers are employed in the occupations most susceptible to task automation. "Nearly two thirds" is that page's wording for the 64.9% of clerical employment. It is not the 84.4% share of the highest-exposure category. Swapping the two numbers invents a different labour market. Inside clerical work, the brief also says female clerks have higher exposure across gradients 1 to 4 (93.5%) than men (87.9%), while men are more concentrated in Gradient 4 (68.4%) than women (61.8%). Both can be true because they use different denominators.12

Sector rates and sector counts also diverge. Shares of employment in gradients 1 to 4 are 82.6% in finance and insurance, 76.3% in wholesale and retail trade, 74.3% in information and communication, 49.4% in transportation and storage, and 47.7% in public administration. Inside Gradient 4, finance and insurance are at 24.2% and public administration at 20.6%. The key points say those two sectors have the largest shares of jobs in the highest category. Wholesale and retail trade accounts for 5.9 million exposed jobs, 51.6% of everything in gradients 1 to 4, because the sector is large, and the brief says most of those jobs sit in the lower gradients. The April news page names financial and insurance services, wholesale and retail trade, and information and communication among the sectors with the highest exposure rates. It does not print 51.6%, 5.9 million, or the public-administration share of Gradient 4.12

Manufacturing is the useful contrast. It employs more than 12 million workers, yet its generative-AI exposure rate is 6.2%, about 0.8 million jobs, including 1.4% in Gradient 4. Most manufacturing workers are in no exposure (82.5%) or minimal exposure (11.3%). The brief says this line is generative AI only. It does not cover industrial robotics or other advanced manufacturing technology. A garment workshop, including the image with this article, is not by itself a clerical Gradient-4 job.1

Geography is a third cut. Ha Noi, Ho Chi Minh City and Da Nang together account for 24.3% of total employment but 38.0% of jobs potentially exposed, more than four million workers. The brief says this reflects the concentration of administrative, professional and service employment in major urban centres. The rates are 36.7% overall and 6.0% in Gradient 4 in Ha Noi, 30.9% and 2.3% in Ho Chi Minh City, and 28.7% and 2.4% in Da Nang. Quang Ninh is next, at 27.4% and 3.4%. The April news page says the three cities account for more than a third of potentially affected jobs. "More than a third" is that page's summary of 38.0%. It is not a claim that other provinces are untouched.12

Urban employment is at 32.9% in gradients 1 to 4 and 3.2% in Gradient 4, against 13.8% and 1.0% in rural areas. In counts, urban areas hold 6.7 million exposed jobs and rural areas 4.8 million. Formal employment is at 28.7% overall and 4.3% in Gradient 4, against 16.9% and 0.6% in informal employment. Among workers with advanced schooling, 42.4% of employment falls in gradients 1 to 4, including 6.6% in Gradient 4. Workers aged 25 to 59 show 22.6% overall exposure. Youth aged 15 to 24 show 15.7%. That age split is not the graduate series in the next section.1

On 8 July 2026 an ILO news page on ASEAN said generative AI is set to touch the working lives of nearly 80 million people. Using estimates the page dates to 2025, 22.9% of ASEAN employment, nearly 80 million workers, is in occupations with more than a minimal degree of potential exposure. Only 3.3%, 11.7 million workers, are in the highest exposure category, and around 67% of employment is in occupations with no identified exposure. Among countries with data, the page lists Singapore at 42.2%, the Philippines at 28.1%, Indonesia at 21.7%, Viet Nam at 20.8% and Thailand at 20.6%. It says there is no evidence to date of large-scale job losses, and that employment in highly exposed occupations has continued to expand. It also says women are more than twice as likely as men to be in occupations with high generative-AI exposure. That "twice" line is the regional comparison for high exposure. It is not Viet Nam's 24.1% against 17.8%. The page does not print a new million-job figure for Viet Nam. The 11.7 million is an ASEAN total, not the around 1.0 million in Viet Nam's Gradient 4. This article did not open the ASEAN brief itself, so it does not treat the repeated 20.8% as a fresh labour-force survey.5

A different Asia series, on hiring into exposed roles, is already discussed in why Asia's AI-exposed hiring gap stopped widening. Those figures are not this exposure index.

A rise in one graduate series is not proof the index was harmless

The brief looks at women and men aged 20 to 24 with advanced education, in four private service sectors with the highest Gradient-4 exposure: information and communication, finance and insurance, professional, scientific and technical activities, and administrative and support services. It says data limits prevent a comparison with studies of young graduates in some advanced economies. It also says 2025 figures were not yet available. From 2022 to 2024, employment in that specific group expanded, by 21.1% in 2023 and a further 16.2% in 2024. The brief's conclusion is narrow: the available survey data do not yet show clear signs of AI-driven hiring declines for recent graduates in those sectors. The April news page says there is no clear evidence so far of declining employment opportunities for young, highly educated workers in high-exposure sectors over 2022-2024, and that employment in highly exposed service sectors has continued to expand. It does not print 21.1% or 16.2%, and it does not limit the sentence to ages 20 to 24 in those four sectors. This article keeps the brief's narrower group. The rise is not a finding that AI created the jobs, and it is not evidence that task change is harmless for everyone else.12

Exposure can also sit beside a longer search when a job does end. That is a different measurement. See when the next job takes longer to find. A job created somewhere else is still not a placement. See why a created job is not a transfer plan.

One factory pilot is a third measurement

On 8 December 2025 the ILO described a pilot at Hanoi Industry Trading Investment Joint Stock Company (HITI), under the Productivity Ecosystems for Decent Work programme. An AI vision system for inspecting mechanical mesh cut inspection time from 20 seconds to 5 seconds, and defect detection from 95% with human inspection only to 99% with human inspection plus AI. The page says the quality-control team was reorganized from 10 workers to 3 senior high-tech supervisors, and that seven workers were retrained and reassigned to higher-value roles. It presents the pilot as strengthening work rather than displacing people. The April brief's box on the same project prints the 20 seconds, the 5 seconds, the 95% and the 99%, and says several workers were retrained and reassigned. It does not print the change from 10 to 3. It says the pilot results are illustrative and not representative of all enterprises. A footnote says the brief is about occupational exposure to generative AI, while the enterprise example involves broader AI applications.16

Those are three different objects: an occupational index, a hiring series for one graduate group, and one named factory's vision tool. Multiplying the time saved on the inspection line by 11.5 million jobs would invent a national result neither paper calculates. Being able to attempt a task with a tool is also not the same as keeping the judgment that supervises it. See who teaches judgment after the first draft and why tasks inside a job can change before the title does.

A national percentage is not a staffing plan

The steps below are an editorial checklist. The ILO did not publish them as a method, and this article does not treat them as a research result.

A 25-person services firm cannot read 20.8% as the share of its own staff to remove. The operational question is which documents, customer conversations, exceptions and decisions its people already perform, which of those tasks are repetitive, and where a tool would add errors, confidentiality problems or a new review. The brief's own warning applies: a low share in the full-automation category is not proof that work will feel the same. A job can keep its title while the valuable tasks, the promotion ladder, the hours and the chance to learn move. A customer-service employee might handle fewer routine questions and more escalations. An administrative assistant might draft less and verify more. Neither change is automatically an improvement. The April news page says the more probable outcome is task transformation, not large-scale displacement. Transformation is not a promise of better work.2

The 2 November page says algorithmic management tools can improve efficiency, while their use without appropriate safeguards can intensify workplace surveillance and pressure. The page says "efficiency." It does not say "coordination." See who oversees the algorithmic manager. The same page says Sinwon Park, Director of the ILO Country Office for Viet Nam, outlined three priorities at Open Innovation Day on 26 October: lifelong learning, investment in people alongside technology, and strong labour market institutions. It also says Sher Singh Verick, the ILO's Global Coordinator on AI and Digitalization, stressed worker involvement in the design and governance of workplace technologies, and social dialogue among governments, employers and workers, at a meeting on 28 October.3

A small organization does not need a national programme to act on that last point. It can ask the people who do the work which tasks are repetitive, which proposed metrics miss rework, and which decisions stay with a person. It should also say what a productivity gain is for: service quality, learning time, shorter turnaround, or more capacity. If training depends on spare time, people with heavier loads or caring responsibilities can be left out. Record access by role and shift, not only course completions. Whether a new skill changes duties, pay or advancement is a separate question from attendance. Who had paid time to practise is already discussed in why a training gap becomes a career gap.

Consider a hypothetical regional charity that uses a tool to sort requests for financial assistance and to draft standard replies. If staff use the saved time to speak with applicants and to resolve complicated cases, the service may improve. If management raises daily quotas without looking at errors or at staff load, the same tool may make the work less sustainable. That scene is an illustration. The ILO did not observe it, and it is not a result of the 20.8% estimate.

For a European organization, a Vietnamese exposure share is not a finding that a system is high-risk. This is not legal advice. Article 4, in the Commission service-desk text of the consolidated AI Act as at 27 July 2026, says providers and deployers shall take measures to support the AI literacy of their staff and of other persons dealing with the operation and use of systems on their behalf. The same paragraph says this obligation does not require a guaranteed level for any individual. The service desk says the summary is not legally binding. A literacy measure is not an instruction to remove staff because an index printed 20.8%.7

Annex III, point 4, is headed "Employment, workers' management and access to self-employment." Point 4(a) lists systems intended for recruitment or selection, including targeted job advertisements, analysing and filtering applications, and evaluating candidates. Point 4(b) lists systems intended to decide terms of a work relationship, promotion or termination, to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate performance and behaviour. An occupational exposure percentage is not on that list. A tool intended to score candidates, to end contracts, or to monitor workers is a different question. Classification depends on the intended purpose. The service desk says Annex III rules apply from 2 December 2027. This article does not decide whether any product is in scope.7

How hilighter can help

hilighter does not estimate exposure in Viet Nam, explain the ILO gradients, or decide whether a workplace tool is high-risk under the AI Act. It can make the skills inside a role visible when a team is deciding which tasks may change and which checks stay with a person. Teams can map skills by role, compare people with those requirements in skills and talent insights, record a check or a correction in skills evidence and recognition, and set personalized growth paths. A manager's review of that evidence sits in manager assessments. The AI job impact view is indicative. It is not an ILO forecast, and it is not a prediction that a role will disappear. The Free plan covers up to three job roles. The number of employees is not capped.

Sources

  1. International Labour Organization, Felix Weidenkaff and Phu Huynh, Generative AI and jobs in Viet Nam: Labour market exposure and policy considerations, ILO Brief, 16 April 2026. Key points: around 20.8% of employment, equivalent to 11.5 million jobs, potentially exposed; the index measures task-level exposure, not adoption or employment effects; 1.8% of jobs, around 1.0 million, in the highest category, where tasks are highly susceptible and relatively standardized. Body: about one in five jobs, 20.8% of total employment or 11.5 million workers, gradients 1 to 4; 1.8% of workers in Gradient 4. Women 24.1% versus men 17.8%; women 55.5% of potentially exposed jobs; Gradient 4 at 2.0% for women; women 52.2% of Gradient-4 jobs; predicted gap after individual and job characteristics of 6.7 percentage points (26.0% versus 19.3%). Clerical employment 64.9% in Gradient 4; clerical workers 0.8 million, or 84.4% of Gradient-4 jobs; female clerks 93.5% in gradients 1 to 4 versus men 87.9%, and men 68.4% in Gradient 4 versus women 61.8%. Sector rates in gradients 1 to 4: finance and insurance 82.6%, wholesale and retail 76.3%, information and communication 74.3%, transportation and storage 49.4%, public administration 47.7%. Gradient 4: finance and insurance 24.2%, public administration 20.6%. Wholesale and retail: 5.9 million exposed jobs, 51.6% of gradients 1 to 4, mostly lower gradients. Manufacturing: more than 12 million workers, 6.2% exposed (about 0.8 million), 1.4% in Gradient 4, 82.5% no exposure, 11.3% minimal; generative AI only, not industrial robotics. Ha Noi, Ho Chi Minh City and Da Nang: 24.3% of employment, 38.0% of potentially exposed jobs, more than four million workers. Rates: Ha Noi 36.7% and 6.0%, Ho Chi Minh City 30.9% and 2.3%, Da Nang 28.7% and 2.4%, Quang Ninh 27.4% and 3.4%. Urban 32.9% and 3.2% versus rural 13.8% and 1.0%; 6.7 million urban exposed jobs and 4.8 million rural. Formal 28.7% and 4.3% versus informal 16.9% and 0.6%. Advanced schooling: 42.4% in gradients 1 to 4, including 6.6% in Gradient 4. Ages 25 to 59: 22.6%. Ages 15 to 24: 15.7%. Ages 20 to 24 with advanced education in four private service sectors: employment up 21.1% in 2023 and 16.2% in 2024; no clear signs yet of AI-driven hiring declines for that group; 2025 not yet available. Index based on Gmyrek et al., ILO Working Paper No. 140 (2025), applied to the 2024 Labour Force Survey; the working paper was not opened separately. The index was built on Poland Labour Force Survey codes and may not fully match local tasks. Box 2 prints the HITI inspection times and accuracy rates, says several workers were retrained, says the pilot is not representative, and does not print the change from 10 workers to 3 supervisors.
  2. International Labour Organization, Generative AI could transform millions of jobs in Viet Nam, new ILO brief shows, 16 April 2026. Around 11.5 million women and men, one in five workers. Task transformation rather than large-scale displacement is described as the more probable outcome. Around one million workers in occupations where tasks are highly susceptible and standardised; full-automation risk less than 2%, lower than in Indonesia, the Philippines and Thailand. Clerical support workers: nearly two thirds in the occupations most susceptible. Finance and insurance, wholesale and retail, and information and communication named among the highest exposure rates. Ha Noi, Ho Chi Minh City and Da Nang: more than a third of potentially affected jobs. Women 24.1% versus men 17.8%, including after education, occupation and other characteristics; the page does not print 6.7 percentage points. Sales: women around two thirds of the workforce. No clear evidence so far of declining opportunities for young, highly educated workers in high-exposure sectors in 2022-2024; employment in highly exposed service sectors continued to expand. The page does not print 21.1% or 16.2%.
  3. International Labour Organization, Making AI work for decent work: ILO calls for human-centred approach as Viet Nam accelerates digital transformation, 2 November 2026. Restates the April estimates: 20.8% of employment, 11.5 million jobs, and around 1.8% in the highest exposure category. Says exposure does not necessarily translate into job displacement. Discussions from 26 to 30 October 2026. No new November labour-force survey. Sinwon Park's three priorities on 26 October: lifelong learning, investment in people alongside technology, and strong labour market institutions. Algorithmic management tools can improve efficiency, and without safeguards can intensify surveillance and pressure. Sher Singh Verick on 28 October: worker involvement in design and governance, and social dialogue. The page does not say "coordination."
  4. International Labour Organization, Workers' exposure to AI: What indicators tell us and what they do not, research brief, PDF headed February 2026. Exposure indicators show technological susceptibility, not labour-market outcomes, and cannot be interpreted as predictions of job displacement, productivity gains or reskilling needs. A general review of indicators, not a Viet Nam count. The April Viet Nam brief's footnote dates a brief of this title to March 2026.
  5. International Labour Organization, AI may affect nearly 80 million workers in the ASEAN region, but large-scale job disruption not yet seen, 8 July 2026. Estimates the page dates to 2025: 22.9% of ASEAN employment, nearly 80 million workers, with more than minimal potential exposure; 3.3%, or 11.7 million, in the highest category; around 67% with no identified exposure. Viet Nam listed at 20.8%, with Singapore 42.2%, the Philippines 28.1%, Indonesia 21.7% and Thailand 20.6%. No evidence to date of large-scale job losses. Women more than twice as likely as men to be in high-exposure occupations across the region, which is not the Viet Nam 24.1% versus 17.8% comparison. The ASEAN brief PDF was not opened. The 11.7 million figure is regional.
  6. International Labour Organization, When AI meets decent work: A productivity breakthrough from the factory floor, 8 December 2025. HITI pilot: inspection time from 20 seconds to 5; accuracy from 95% (human only) to 99% (human plus AI); quality control reorganized from 10 workers to 3 senior supervisors; seven workers retrained and reassigned. The page presents this pilot as not displacing workers. One enterprise, not the 11.5 million exposure figure.
  7. European Commission, AI Act Service Desk, Article 4: AI literacy and Annex III, points 4(a) and 4(b). Wording reflects the EUR-Lex consolidated version as at 27 July 2026. The summary is not legally binding. Paragraph 1 of Article 4 does not require a guaranteed level of literacy for any individual. When does enforcement start?: Annex III rules from 2 December 2027. Not a decision about any product. Not legal advice.

Sources last reviewed: 17 November 2026. The 11.5 million figure is not treated as a layoff count. "Jobs" in the key points and "workers" in the body are reported as the brief prints them. The April news line "less than 2%" is not swapped with the brief's 1.8%. The clerical 64.9% is not swapped with the 84.4% share of Gradient-4 jobs. The ASEAN "more than twice" line for women is not merged with 24.1% versus 17.8%. The 21.1% and 16.2% rises are not treated as caused by AI, and they are not applied outside ages 20 to 24 with advanced education in the four named sectors. The HITI 10-to-3 change is kept on the December page and is not read back into the exposure index. The charity scene is an illustration. EUR-Lex was not opened separately. Microsoft adoption rates, the 81% daily-use line, and Decision No. 433/QD-TTg appear in the April brief and were not used, because those underlying pages were not opened. Convention No. 193 is named on the November page and is not interpreted here. Neither an AI-detector score nor the absence of Google penalties can be guaranteed.