NEW REPORT: Artificial Intelligence Is Amplifying Gender Inequality Across Nearly Every Aspect of Women’s Lives
Landmark analysis from Gender Equity Policy Institute finds AI systematically disadvantages women in hiring, lending, healthcare, and criminal justice; warns that biased outputs can feed future AI models, compounding discrimination over time.
According to a landmark new report from Gender Equity Policy Institute (GEPI), titled How AI Fails Women: When Bias Becomes Amplified, artificial intelligence is systematically reinforcing gender inequality, putting women at a disadvantage in employment, financial services, healthcare, and the criminal justice system.
Authored by machine learning scientist and GEPI senior fellow Dr. Myriam Shiran, the report draws on an extensive body of existing research and independent audits to examine how AI systems perpetuate discrimination against women. While researchers have documented gender bias in individual uses of AI technology for years, GEPI’s analysis brings those findings together across six major areas of women’s lives, revealing a broader and more troubling pattern.
The report finds that AI systems are learning from decades of unequal treatment of women and reproducing those biases in decisions that can have significant consequences for women today. It explains how those biases become entrenched and perpetuated over time, as biased outputs influence real-world decisions and become part of the data used to train future systems.
Taken together, GEPI’s findings raise serious concerns about the growing use of AI in American institutions, and whether existing safeguards are sufficient to prevent the technology from making gender inequality worse.
“Women have fought for generations to overcome discrimination and win rights and opportunities that were once denied to us. Yet we’re now building AI systems using data shaped by that very history of unequal treatment,” said Dr. Nancy L. Cohen, founder and CEO of the Gender Equity Policy Institute. “For decades, medical research treated men as the default, hiring and lending practices systematically disadvantaged women. Now that same legacy of inequality is being carried into the algorithms making decisions about our lives. Without meaningful safeguards, AI risks cementing the discrimination of the past into our present reality, and making it even harder to overcome in the future.”
Among the report’s key findings
- AI hiring tools can favour men, even when candidates have identical qualifications. In a study involving more than three million simulated résumé comparisons, researchers found that AI screening models significantly favoured résumés with male-associated names in 51.9% of the tests, compared with 11.1% favouring female-associated names. Other research found that removing explicit gender identifiers often fails to eliminate discrimination because algorithms can infer gender through work histories, education, and other details.
- Medical AI can recommend different care for identical patients based on gender. In one study, GPT-4 rated cardiac stress testing as highly important in 71% of repeated evaluations of a male patient, compared with just 58% involving an otherwise identical female patient. Cardiologists presented with the same clinical scenario in an earlier study showed no statistically significant gender difference in recommending the test. In another experiment, GPT-4 ranked panic or anxiety disorder higher for a female patient whose symptoms were actually caused by a potentially fatal pulmonary embolism.
- AI-driven lending systems can charge women more than their financial risk warrants. Research examining approximately 80,000 loans from a major U.S. fintech lender found that an algorithm systematically overestimated women’s likelihood of defaulting, resulting in higher borrowing costs than their actual repayment risk justified. The lender earned higher returns on loans to women because of the discrepancy.
- Criminal justice risk-assessment tools can systematically overestimate women’s likelihood of reoffending. Among federal offenders assigned the same moderate-risk classification, 52% of men were rearrested compared with 37% of women. The findings show how tools developed largely around men’s patterns of offending can misclassify women, potentially exposing them to unnecessarily restrictive supervision and sanctions.
- Generative AI reinforces stereotypes about women’s abilities and roles. Studies found that AI-generated recommendation letters emphasised men’s leadership, expertise, and intelligence while describing women in terms of warmth, emotion, and appearance. Image generators similarly portrayed high-status occupations as disproportionately male. Separate research found that exposure to stereotyped AI-generated images increased viewers’ own gender biases.
- AI is enabling new forms of sexual abuse disproportionately targeting women and girls. Consumer applications can create fabricated sexual images of real people using ordinary photographs. An analysis of 134 Reddit discussions about nonconsensual deepfake pornography found that 85.8% of victims described were women, while 81.3% of perpetrators described were men. Separate research examining requests on a major AI-image marketplace found that women were the targets of nearly 90% of all deepfake commissions involving real people, including 23 of 24 sexually explicit deepfake requests.
The report identifies four recurring ways that AI systems perpetuate discrimination. They learn from historically unequal data, reproduce established gender stereotypes, and present their outputs as objective assessments. Those outputs can then become part of the information used to train future systems, allowing the bias to persist and compound.
“Researchers have been documenting the ways AI disadvantages women for years. What this report shows is how deeply connected those problems are. Across hiring, lending, healthcare, and criminal justice, we see the same underlying failures. These systems learn from records shaped by discrimination and turn that information into decisions that are presented as objective. Even removing gender from the data doesn’t necessarily eliminate the bias. We’re dealing with a systemic problem that goes far beyond any one algorithm or industry.”
Dr. Myriam Shiran, report author and Senior Fellow at Gender Equity Policy Institute
By bringing together research across six areas of women’s lives, the report identifies broader patterns that have largely gone unaddressed as AI becomes more widely adopted. It finds that current oversight often fails to account for how gender bias operates across different systems or how discriminatory outcomes can compound over time. Companies frequently determine the scope of their own bias assessments, and some automated fairness tests have failed to identify significant differences in how AI systems treat women and men.
Without stronger oversight, GEPI warns, these technologies risk making existing inequalities more entrenched and increasingly difficult to address.
The report concludes that addressing gender bias in AI will require more than correcting individual algorithms. GEPI makes the case that stronger oversight is needed to ensure these systems are properly tested, independently scrutinised, and held accountable for discriminatory outcomes. As AI becomes more deeply embedded in American life, the consequences of failing to address these problems will only grow. A forthcoming report in GEPI’s series on gender and AI will outline specific regulatory principles and policy recommendations to address these systemic failures.
About Gender Equity Policy Institute
Gender Equity Policy Institute (GEPI) is an independent research institute dedicated to advancing women’s freedom, well-being, and opportunity. GEPI specialises in data-driven, historically informed research and policy development on the most pressing challenges facing Americans. Its work applies a gender equality perspective to public policy, with the goal of closing gender gaps and expanding opportunity.
