Artificial intelligence was supposed to make work fairer. New data suggest it may do the opposite.
When Anne, a data entry clerk, watched an IBM PC arrive on her desk in 1986, her job was gone within a year. Four decades on, a social-media manager watches ChatGPT draft the posts she once wrote. The difference, as Noreena Hertz observed last year in Project Syndicate, is that this time the exit may come faster. And it is falling, with uncomfortable regularity, on women.
The gender pay gap has survived recessions, policy reforms and decades of corporate diversity pledges. What is less well understood is how artificial intelligence is reshaping the terrain in ways that could make it worse. Not through overt discrimination, but through the collision of two forces: where women work, and how reluctant many are to use the technology now rewarding those who do.
Pink-collar exposure
A research brief published in May 2025 by the International Labour Organisation found that female-dominated occupations are almost twice as likely to be exposed to generative AI as male-dominated ones. Around 29 per cent of female-dominated roles face significant disruption, against 16 per cent of male-dominated ones, and the exposure gap widens sharply at the high-risk end, where 16 per cent of female-dominated occupations fall into the most vulnerable categories, against just three per cent of male-dominated ones.
The reasons are structural. Women are heavily concentrated in clerical and administrative roles (secretaries, payroll clerks, receptionists, accounting assistants) where tasks are routine, codifiable and exactly the kind of work at which generative AI excels.
Men, by contrast, are disproportionately found in construction, manufacturing and manual trades, where tasks resist easy automation. In Europe and Central Asia, 39 per cent of women’s jobs are exposed to AI disruption, compared with 26 per cent of men’s. In high-income countries globally, that figure climbs to 41 per cent for women and 28 per cent for men.
The World Economic Forum and LinkedIn, in a white paper published in March 2025, put it bluntly: relatively fewer women are in jobs being augmented by AI, and relatively more are in jobs being disrupted by it. In the United States, 24.1 per cent of men work in augmented occupations; 20.5 per cent of women do. At the other end, 33.7 per cent of women work in roles being disrupted, against 25.5 per cent of men. The pattern holds in 95 per cent of the 74 countries surveyed.
The adoption gap
Displacement, though, is only half the story. The other risk is falling behind on skills that are fast becoming a condition of higher pay. Research from Wharton published in late 2025 found that access to new technologies such as AI, cloud systems and their equivalents, is now one of the biggest drivers of earnings in the tech sector. Because fewer women are working with these tools, the access gap is widening the earnings gap. Workers with demonstrable AI skills, according to PwC’s 2025 AI Jobs Barometer, earn on average 25 per cent more than peers without them.
The trouble is that women are not acquiring those skills at the same rate as men. A meta-analysis from researchers at Harvard and Berkeley, synthesising data from 18 studies covering around 143,000 people worldwide, found a 25 per cent gap in generative AI adoption between men and women. The Federal Reserve Bank of New York’s Survey of Consumer Expectations, conducted in early 2024, showed that 50 per cent of men had used generative AI in the previous 12 months, against 37 per cent of women. Between November 2022 and May 2024, women made up only 42 per cent of ChatGPT’s average monthly website users; when it came to smartphone app downloads, that share fell to 27 per cent.
The gap persists even when access is equalised. When researchers at Harvard invited some 17,000 male and female entrepreneurs in Kenya to use ChatGPT by providing free access and instructions, women were still around 13 per cent less likely to engage with the tool. “Even when the opportunity to use ChatGPT was equalized, women were less likely to engage, which we think is pretty shocking,” said Rembrand Koning of Harvard Business School. The gap is not simply one of access.
Part of the explanation lies in risk perception. Women were more likely to see AI use as a form of cheating and more worried about professional penalties for relying on the technology. “Women face greater penalties in being judged as not having expertise in different fields,” Koning noted. Research from UC Berkeley confirmed this: women reported lower confidence in their ability to use gen AI tools and were more likely to view adoption as professionally risky.
Biased by design
There is a third dimension that receives less attention than it deserves. AI is not merely reshaping who does what work, but it is also, in a growing number of organisations, making decisions about who gets paid what and who gets hired at all. The results, several researchers warn, are far from gender-neutral.
When AI tools are trained on historical employment and compensation data (data that already reflects decades of occupational segregation and pay discrimination) they tend to reproduce and sometimes amplify the patterns embedded in that data.
Pay-benchmarking algorithms may recommend lower salary ranges for roles historically dominated by women. Performance-scoring systems trained on male-coded definitions of leadership may systematically give women lower ratings. Hiring tools that penalise non-linear career paths (disproportionately the result of caring responsibilities) will disadvantage women at the point of entry. The EU’s AI Act, which designates AI used in employment as ‘high risk’ and requires conformity assessments on gender-disaggregated data, acknowledges the problem explicitly. Enforcement is another matter.
Reasons for qualified optimism
The same WEF and LinkedIn analysis also noted, however, that women’s representation among AI engineering skill-listers on LinkedIn rose from 23.5 per cent in 2018 to 29.4 per cent by early 2025, with the gap narrowing in 74 of 75 economies surveyed. The scarcity of AI talent may push employers to cast their recruitment nets wider. And the ILO, for its part, emphasises that outright job destruction is less likely than transformation — that most AI-exposed roles will change rather than disappear.
But the window for course correction is narrowing. The productivity premium on AI skills is already visible in wage data; the longer the adoption gap persists, the wider the pay gap it feeds. Randstad’s Workmonitor 2025 found that AI skilling was a top-three priority for 44 per cent of men globally, against 36 per cent of women. Among AI talent overall, 71 per cent are men.
For employers and governments in Central and Eastern Europe and beyond, there are lessons to be learnt. The region has made genuine strides in female workforce participation and narrowing pay gaps over recent decades. AI, left to its own devices, threatens to undo some of that progress with an efficiency that no glass ceiling ever managed.
Photo: Dreamstime.