AI Is Turning Old Gender Bias Into an Urgent Global Risk

UN Women has warned that artificial intelligence systems are reproducing gender stereotypes, amplifying online abuse and excluding women from decisions shaping the digital future. The warning comes ahead of major UN-linked AI governance meetings in July and reflects a broader concern: as generative AI becomes embedded in work, media and marketing, bias can move from isolated outputs into systems used at scale.

Artificial intelligence (AI) is already shaping how people work, communicate, advertise, search for information and experience public life. Meanwhile, UN Women has warned that the technology is reproducing stereotypes, intensifying online violence and excluding women from the decisions that will shape the digital future.

The danger is not that AI may one day become biased. The evidence suggests the bias is already present and the challenge now is whether governments and companies can respond before it is built into systems used at scale.

The Bias Is Being Automated

The evidence cited by UN Women suggests gender bias in AI is not a rare malfunction. A study of 133 AI systems found that 44 per cent demonstrated gender bias, while 26 per cent demonstrated both gender and racial bias.

Large language models have also been found to associate women with words such as "home," "family" and "children," while linking men with "business," "executive," "salary" and "career." In some sentence-completion tasks, about 20 per cent of model responses showed sexist or misogynistic attitudes, including portrayals of women as sex objects or as property of their husbands.

Generative AI is now part of everyday content creation. In the United Kingdom, 88 per cent of advertising and media agencies are already using generative AI in some form. Yet only 51 per cent of marketers use human oversight to test AI-generated creative before release. When biased systems are used casually or at scale, stereotypes can move from isolated outputs into advertising campaigns, workplace messaging, media content and public-facing communications.

The concern is no longer limited to whether AI can produce biased content, but whether institutions are checking for that bias before it reaches audiences.

Online Abuse Is Entering a More Dangerous Phase

UN Women also warns that AI is intensifying violence against women and girls in digital spaces. Women human rights defenders, activists and journalists are already frequent targets of online harassment. According to the data, almost one in four surveyed women in these groups had experienced AI-assisted online violence. Twelve per cent reported the non-consensual sharing of personal images, including intimate or sexual content, while six per cent said they had been targeted through deepfakes or manipulated images and videos.

Deepfakes show how quickly AI can turn existing abuse into a more scalable threat. The technology can be used to create realistic fabricated images or videos, making harassment easier to produce and harder to contain. For women in public life, journalism, activism or politics, the damage can be personal, professional and lasting.

This is where AI governance becomes more than a technical debate. The spread of synthetic content raises legal, platform and enforcement questions that many institutions are still struggling to answer. Who is responsible when manipulated images circulate? How quickly must platforms respond? What protections exist for victims? And how can accountability keep pace with tools that are becoming cheaper and more accessible?

Women Are Still Missing From the Rooms Where AI Is Built

UN Women's warning also points to a structural problem: women remain underrepresented in the AI workforce. They make up only 30 per cent of the global AI workforce, even as generative AI is expected to drive job growth in technology-intensive sectors.

The imbalance is significant because AI systems are shaped by human decisions. Choices about training data, safety testing, product design, moderation, deployment and oversight all influence how these systems behave. If the people building and governing AI do not reflect the communities affected by it, some harms are more likely to be overlooked or addressed too late.

The labour-market impact may also be uneven. Women outside the AI sector are nearly twice as likely as men to hold jobs at high risk of automation. That does not mean automation will affect all women workers in the same way, or that job losses are inevitable. But it does raise a policy challenge: AI could create opportunities in some sectors while increasing insecurity in others.

The impact is likely to be shaped by more than gender alone. Race, disability, income, geography and digital access can all deepen exposure to harm or exclusion. It makes gender-responsive AI governance not a narrow equality demand, but a broader test of whether technological change is being managed fairly.

The Business Case Is Becoming Harder to Ignore

The Unstereotype Alliance, convened by UN Women, has cited global research showing that inclusive advertising is linked to stronger business outcomes. Inclusive advertising, free of gender stereotypes, was associated with a 3.46 per cent short-term sales uplift and a 16.26 per cent long-term sales uplift. It was also linked to higher brand preference, stronger pricing power and greater customer loyalty.

As brands increasingly use AI to plan, write, design and distribute campaigns, those findings carry practical significance. Companies that fail to check AI-generated content for bias may face reputational damage, weaker audience trust and avoidable commercial risk. Those building inclusion into their AI workflows may be better positioned to reach wider audiences and avoid harmful stereotypes.

The Unstereotype Alliance playbook launched in June 2026 gives marketers a tool for identifying bias before AI-generated content is released. Its impact will depend on whether brands, agencies and technology vendors treat it as a routine part of production rather than an optional reputational safeguard.

The larger governance question remains unresolved. Of 138 countries assessed, only 24 referenced gender in a national AI strategy, and just 18 included substantive gender-responsive provisions, suggesting that many governments are still treating AI primarily through the lenses of innovation, productivity, competitiveness and safety, while gender equality remains underdeveloped in policy design.

The upcoming United Nations Global Dialogue on Artificial Intelligence Governance and the AI for Good Global Summit in Geneva will therefore be watched not only for broad statements about responsible technology, but for signs of practical action. Will governments strengthen national AI strategies with gender-responsive provisions? Will companies make human review of AI-generated content standard practice? Will platforms and regulators move faster on deepfakes and image-based abuse? Will women and girls, including those from marginalized communities, have a meaningful role in shaping AI rules?

AI can help detect stereotypes, improve accessibility and broaden representation, but those outcomes are not automatic. They depend on who designs the systems, what data they learn from, how they are governed and whether institutions are willing to slow down deployment long enough to test for harm. The technology can either repeat old inequalities at machine speed, or it can be governed with enough care to avoid locking them into the digital future. The window for making that choice is narrowing.