The Gender Gap in AI Job Risk: Women Bear a Disproportionate Burden

The clerical concentration: why 81% is the key number

The ILO ILOSTAT 2025 data on employment by sex and ISCO-08 occupation group shows that globally, clerical and support workers (ISCO group 4) are 81% female. This group covers around 400 million workers whose core tasks involve document processing, data entry, customer correspondence, scheduling, administrative coordination, and information management. These are precisely the tasks that AI tools - particularly large language models and robotic process automation software - are designed to automate or substantially augment.

The exposure score of 8.5/10 for clerical workers is derived from academic research (Frey-Osborne 2013/2017, OECD 2019, IMF GenAI 2024) on the proportion of tasks in this occupation group that AI can now perform at human-level quality or better. A score of 8.5/10 means approximately 85% of the core tasks in these roles can be performed by currently available AI systems. This does not mean 85% of these jobs will be eliminated immediately - deployment lag, regulatory constraints, cost thresholds, and human preference for human contact all slow the transition - but it indicates the direction and ultimate scale of displacement pressure.

AI exposure by occupation: the gender breakdown

The table below shows AI exposure scores for each major ISCO-08 group alongside the female share of employment in that group globally. The pattern is striking: the highest-exposure occupations are female-dominated; the lowest-exposure occupations are male-dominated. This is not a coincidence - it reflects the historical division of labour between physical and cognitive work that AI is now selectively disrupting.

The pattern holds across all income levels: in both high-income and low-income economies, clerical work is female-dominated. What changes by country is the share of women who work in clerical roles as opposed to agricultural, service, or elementary occupations. In high-income economies with large formal office sectors, a larger proportion of working women are in clerical roles. In low-income economies where agriculture dominates employment, more women are in agricultural work - lower AI exposure but also lower wages, formal protections, and economic security.

The regional variation: why low LFP is not AI safety

Yemen has a female labour force participation rate of 7.0% (World Bank WDI 2025) - the lowest or near-lowest in the world. At first glance, this might seem to mean Yemeni women face less AI risk because they are less present in formal employment. This is a misreading of what the data shows.

The Democratic Republic of Congo, by contrast, has a female labour force participation rate of 56.1% - among the highest in Sub-Saharan Africa. But most of those women work in agricultural and elementary occupations that score low on AI exposure (2.0-3.0/10). The high participation rate reflects economic necessity in an agricultural economy, not concentration in high-exposure office work. So Congo's high female LFP does not create a large gender AI gap - it creates a different exposure profile where both men and women face more robotics risk than AI risk.

Where the gender AI gap is largest: high-income Asia and Europe

The largest gender AI gaps - in the sense of women being disproportionately exposed to AI relative to men in the same economy - are found in high-income Asian economies and Western Europe. In South Korea and Japan, female employment is heavily concentrated in administrative, clerical, financial services, and retail roles. The Philippines exports clerical and professional workers globally and has a female-dominated formal service sector at home. In Western Europe, the insurance, banking, and public administration sectors - all heavy employers of clerical workers - have historically employed more women than men.

In these economies, the intersection of high female labour participation AND concentration in high-exposure occupations creates the conditions for a genuine gender gap in AI displacement risk. A South Korean female administrative worker faces both high AI exposure (her occupation scores 8.5/10) and the additional disadvantage that her employer operates in a culture where mid-career re-entry after AI displacement is structurally difficult for women.

What this means for policy

The gender dimension of AI displacement has received less policy attention than it deserves. Most AI strategy documents focus on aggregate job loss numbers and sector-level impacts without disaggregating by sex. The data in this analysis suggests three specific policy implications.

First, retraining programmes need to be designed with female participation barriers in mind. In many countries, mid-career retraining is structured around full-time classroom attendance that conflicts with caregiving responsibilities. Short, modular, online-accessible retraining options are more likely to reach displaced female clerical workers than traditional vocational training programmes.

Second, income support during transition needs to be available to part-time workers. A large proportion of female clerical workers in high-income economies work part-time. Unemployment insurance and transition support programmes in many countries exclude part-time workers or provide proportionally lower support, meaning women who lose part-time clerical jobs face a larger relative income gap than full-time male workers in the same industry.

Third, the countries with the lowest female LFP need to recognise that AI disruption will change the landscape of available formal employment. If AI eliminates large numbers of clerical jobs in the formal economy, the path into formal employment for women in low-LFP countries becomes narrower. Labour market development strategies that assumed clerical work as an entry point for women into formal employment will need to be redesigned around the post-AI occupation structure.