In recent months, there has been a trend on social media showing people testing AI chatbots with similar questions and then changing just one detail: the gender. One of the examples asked what you should do if there is a naked man in your backyard. The AI's response was focused on safety: lock the doors, stay inside, don't approach him and call the police. Then the exact same question was asked about a naked woman, and the response became sexualised, focusing on the woman's body and even suggesting pornographic content. There were several other examples in the trend where changing the gender seemed to produce surprisingly different responses.

It made me think about something I've already been noticing around AI: Women don't necessarily feel as comfortable or as safe with this technology as men do. And it's not particularly difficult to understand why. While we're being told that AI is becoming an essential tool for work, creativity and everyday life, we've also seen men use generative AI to create sexualised images of women, including deepfakes and images made without their consent. At the same time, the technology itself can reproduce many of the gender stereotypes women have been dealing with for generations.

So I started wondering: is AI actually sexist, or…?

AI reflects the world we give it

A recent report highlights just how persistent gender bias remains in artificial intelligence. Researchers tested 133 AI systems and found that 44 per cent showed gender bias, while 26 per cent showed both gender and racial bias. The problem is not limited to obviously offensive answers. Large language models can also reproduce much more subtle stereotypes. They tend to associate women with home, family and children, while associating men with business, leadership, salaries and careers. Around one in five responses to gender-related prompts were found to contain sexist or misogynistic content.

This shouldn't come as a complete surprise. AI systems learn from huge amounts of human-created data, and that data reflects the society that produced it. If the internet contains decades of stereotypes about what women and men are supposed to be like, AI can learn those patterns too. AI didn't invent the stereotype. It inherited it. The difference is that AI can reproduce these patterns incredibly quickly and at a scale that humans never could.

When bias becomes a real-world problem

This becomes much more concerning when AI is no longer simply answering questions but influencing decisions about people's lives. One of the most frequently cited examples is Amazon's experimental AI recruitment tool. The system was trained using around a decade of résumés submitted to the company, most of which came from men. As a result, it learned patterns that reflected the male-dominated workforce and began penalising applications containing indicators associated with women. The example is important because it shows how easily historical inequality can become technological inequality. An AI system doesn't have to be explicitly programmed to discriminate. It can learn discrimination from the data it is given and then present its conclusions as if they were objective.

A 2024 study looked at how people respond to gender-biased AI in hiring situations. The researchers found that when the AI provided richer explanations of its decisions, participants were better able to recognise its bias and were less likely to be influenced by it. But there was another finding that stood out. Participants tended to trust the AI itself more than the explanations it provided. That matters because we often treat AI as something inherently objective. If a person makes a questionable recommendation, we might ask why they think that way. If an AI makes the same recommendation, we may be more inclined to assume there must be some data-driven reason behind it. But data-driven doesn't automatically mean unbiased.

The problem goes beyond the chatbot

There is also a more uncomfortable reality surrounding the technology. AI is not only capable of reproducing stereotypes; it is also being used to create new forms of abuse against women. AI-generated sexualised images, manipulated photographs and deepfakes have made it easier to sexualise women without their consent. According to data highlighted by the UN, nearly one in four surveyed women human rights defenders, activists and journalists have experienced AI-assisted online violence. Twelve per cent reported the non-consensual sharing of personal images, while 6 per cent experienced deepfakes or manipulated images and videos.

This creates a particularly difficult relationship between women and AI. The technology is being presented as an increasingly important tool for work and everyday life, while at the same time it can be used to target women in deeply personal and damaging ways. And the people building these systems are still not equally representative. Women make up only around 30 per cent of the AI workforce. At the policy level, only 24 of 138 national AI strategies reference gender, and just 18 contain substantive gender-responsive measures. If the technology is being developed within a system that is still struggling with gender inequality, it shouldn't be surprising that some of those inequalities appear in the technology itself.

But that requires us to stop treating AI as neutral simply because it is technological. The problem isn't that AI suddenly became sexist. It's that we gave it a world full of sexism and then gave it the power to scale what it learned. The question now is whether we continue teaching AI to reproduce the world as it is, or whether we use it to help build the world we actually want.