Gender Bias Found in 44% of AI Systems as Adoption Accelerates

Bias embedded at scale

A comprehensive study examining 133 artificial intelligence systems has uncovered troubling patterns: 44 percent demonstrated gender bias, while 26 percent exhibited both gender and racial bias. The findings arrive as generative AI becomes standard practice in marketing and communications, with 88 percent of UK advertising and media agencies already deploying the technology in some capacity.

The concern extends beyond prevalence to oversight. Only 51 percent of marketers currently employ human review to test AI-generated creative before publication, according to data released by UN Women ahead of major AI governance summits in Geneva this July.

Why it matters

As AI systems make decisions about content creation and media buying at unprecedented scale and speed, discriminatory algorithms risk amplifying existing inequalities across billions of interactions. The window to establish guardrails is narrowing as adoption accelerates faster than policy frameworks can respond.

Patterns in language models

Large Language Models consistently associate women with domestic concepts like "home," "family," and "children," while linking men to professional terms including "business," "executive," "salary," and "career." When prompted to complete sentences beginning with gender identifiers, approximately 20 percent of LLM responses exhibited sexist attitudes, including portrayals of women as sex objects or property.

These outputs reflect training data drawn from decades of unequal representation. Of 138 countries assessed, only 24 reference gender in national AI strategies, and just 18 include substantive gender-responsive provisions.

Violence and economic displacement

AI-enabled abuse disproportionately targets women and girls. Nearly one in four surveyed women human rights defenders, activists, and journalists reported experiencing AI-assisted online violence. Twelve percent experienced non-consensual sharing of personal images, while six percent were targeted through deepfakes or manipulated media.

The economic disruption poses additional risks. Women outside the AI sector face nearly double the likelihood of holding jobs at high automation risk compared to men. Meanwhile, women comprise only 30 percent of the global AI workforce, creating a representation gap in the teams building these systems.

Commercial case for inclusion

Research from the Unstereotype Alliance, an industry initiative convened by UN Women, demonstrates that inclusive advertising delivers measurable business value. Brands creating content free of gender stereotypes achieve 3.46 percent higher short-term sales and 16.26 percent higher long-term sales. These brands are 62 percent more likely to become a consumer's first choice and experience 54 percent higher pricing power.

The Alliance launched a playbook in June 2025 providing marketers with frameworks to identify bias before deployment when using generative AI tools.

UN Women emphasizes that when designed with safety protocols and deployed intentionally, AI can help detect stereotypes, broaden representation, and improve accessibility at scale. Achieving these outcomes requires incorporating voices and lived experiences of women and girls from diverse contexts throughout the AI lifecycle—from development through deployment and governance.

These findings were first reported by UN Women in connection with the United Nations Global Dialogue on Artificial Intelligence Governance and AI for Good Global Summit.