AI hiring tools may be keeping women out of work

Amazon scrapped a secret recruiting system after machine-learning specialists found it did not like women, and the model had been trained on a decade of resumes that reflected a largely male applicant pool. That is the core risk now facing women who step out of work for caregiving and later try to return: the system may read a non-linear career as weakness instead of context.

How the penalty gets built into hiring software

The Amazon case became a warning because the tool learned from past hiring patterns rather than from a clean definition of merit. Resumes that contained terms linked to women’s activities were downgraded, which showed how historical imbalance can be converted into a scoring rule.

That logic matters far beyond one company. A July 22, 2024 piece in the Gender Policy Report by Sophie Kniepkamp, Florian Pethig and Julia Kroenung kept the focus on algorithmic bias in job hiring, while Cardozo Law Review scholars Lori Andrews and Hannah Bucher argued that people underestimate the effect algorithms and recommendation engines have on jobs.

Why women returning to work are especially exposed

Career breaks, part-time roles and sideways moves are common in real labor markets, especially for women who have spent time on childcare or eldercare. AI hiring systems often reward neat chronology, long uninterrupted tenure and keyword patterns that resemble previous successful applicants. That means a resume with gaps can be treated as a risk signal even when the gap reflects caregiving rather than disengagement.

This is why return-to-work candidates can be hit twice. First, the model may rank them lower because it has learned from historically male-dominated patterns. Second, the data it uses may assume that the best worker is the person whose resume looks the most conventional. When an algorithm is trained on past hiring decisions, it can inherit the idea that a straight career line is normal and everything else is deviation.

The International Labour Organization has a policy dialogue on “Artificial Intelligence (AI) and non-discrimination in the world of work.” UN Women argues that AI is getting women wrong and that gender equality has to be built into the systems themselves.

What the latest data say about exposure

The OECD’s December 6, 2024 report, “Algorithm and Eve: How AI will impact women at work,” found that women and men face roughly the same overall occupational exposure to AI, but women are underrepresented in the occupations with the very highest exposure to AI. Clerical occupations are highly exposed to AI, and women are overrepresented there, especially women without tertiary education.

Hiring software does not operate in a vacuum. The OECD also found that employment growth between 2012 and 2022 was more rapid in the occupations most exposed to AI. So the jobs that are expanding are also the jobs where automation pressure is strongest, and women are not evenly distributed across those roles.

UN DESA put a sharper number on the broader risk on September 18, 2024: 3.7% of jobs held by women are likely to be replaced by AI, compared with 1.4% of jobs held by men.

Why audits matter, and what employers need to look at

On April 3, 2024, the University of Essex wrote that there is “no quick tech fix” for AI bias against women job applicants. Bias does not disappear when software is added. It has to be measured, challenged and documented.

A serious audit begins with the inputs, not the branding. Employers need to know what signals their systems are trained on: past hires, rejected applicants, keyword matches, education patterns, employment gaps, prior titles and tenure. If a model learns that uninterrupted full-time work is the safest route to a hire, then part-time experience and caregiving breaks become invisible costs.

A credible audit also tests outputs by gender and by career pattern. That means checking whether women with gaps are ranked lower than men with similar qualifications, whether part-time roles are weighted as less valuable than equivalent experience, and whether a resume changes score because of wording associated with caregiving, schools, volunteer work or women’s organizations.

The accountability question is now institutional

Dr Emily Yarrow at Newcastle University framed the issue in 2025 in broader terms, asking whether AI can be trusted to hire fairly and inclusively at a time when algorithms increasingly shape daily decisions. If companies cannot explain what their systems reward, they cannot prove those systems are neutral.