Why female AI agents got paid less than male ones in a new study

Imagine two coworkers doing exactly the same work on exactly the same software — and one getting paid noticeably less, for no reason except a name and a face. That is what happened to female AI agents in a 2026 workplace experiment: 189 workers inside a virtual-reality office handed the female-presenting agent Johanna 10.25% less real money than the male-presenting Johan for completing identical tasks, according to the University of Limerick research announcement. The twist is that both agents ran on the same underlying AI. The difference was never in the technology — it was in the people watching it.

Researchers from the University of Zurich, the University of Limerick and SKEMA Business School built the experiment to capture behavior rather than opinions. Participants worked through office tasks with four different AI assistants: a plain text chatbot, a desk robot, and two human-like virtual agents — Johan and Johanna. Every assistant used the same underlying language model from OpenAI, so any gap in treatment could only come from how each agent was presented. At the end of each session, workers received real cash and divided it between themselves and the AI assistant, turning an abstract question about bias into a measurable transaction, as reported by Cybernews. The central question was simple: would female AI agents earn the same as male ones when the work was identical?

How female AI agents got paid less for the same work

On paper, the two human-like agents were judged almost equally. Participants rated Johan and Johanna similarly for the quality of their work, with Johan trusted only a touch more — a difference too small to be statistically meaningful. But when real money changed hands, both men and women gave Johanna meaningfully less than Johan. Johan also came across as more human-like than his female counterpart, a perception gap the researchers say is itself worth unpacking.

The most revealing part came after the money was counted. In 34 follow-up interviews, most participants insisted that the gender of an AI assistant did not matter to them and that they had no preference between male and female AI agents — yet their payments said otherwise. One participant even told researchers that having Johanna beside them made them "more prone to credit her more," because it felt like a real colleague was sitting there sharing ideas. What people said and what they did, it seems, are two different things when bias is running quietly in the background.

Why female AI agents reveal a human problem, not a robot one

To be clear, this study is not proof that AI itself is sexist. It is evidence that humans project familiar social biases onto AI presentation — and the researchers are the first to stress the limits. This was one controlled experiment with fewer than two hundred participants, in one virtual office, using one family of AI models. It does not prove a universal pay gap across every workplace, and the fact that Johan was rated more human-like leaves room for competing explanations. What it does prove is that a gendered name and avatar can move real money, even when everyone involved swears gender played no role.

That should sound familiar, because the human version of this gap is still very real. According to an HR Review analysis of UK payroll data, the median gender pay gap in IT has widened to 17.6%, its highest level in nearly a decade, driven by fast-growing AI roles where men are heavily overrepresented. LinkedIn research found women made up only about a quarter of new hires into AI positions over the past year, compared with half of new hires in non-AI technical roles, as reported by RocketNews. If the people building AI systems skew heavily male while the assistants serving us skew female, female AI agents are not escaping the pattern — they are inheriting it.

There is precedent for this. For years, the default voices of major AI assistants were female, a design choice critics said reinforced the idea of women as helpers and secretaries. Now the same design levers — voice, name, avatar — are deciding which female AI agents get trusted, credited and paid. The bias did not come from the model; it came from the packaging, and the packaging is a human choice.

What female AI agents mean for your working life

This is not abstract for anyone starting a career right now. AI agents are already running job interviews, customer-service chats increasingly have names and faces, and AI influencers now front real brand campaigns. As these systems shift from anonymous tools to digital coworkers, every design choice becomes a workplace decision. A female-sounding support agent that customers rate lower, or a male-named coding assistant that gets more credit, can quietly recreate the old pay gaps inside new software.

So what do you do with that knowledge? Start by noticing presentation when you evaluate an AI at work — or when your employer does. Push for objective measures wherever possible: accuracy, completion rates, resolution times, revenue influenced. Those numbers are harder to nudge with a name or an avatar than a star rating is. And when a company rolls out AI coworkers, ask whether it has tested how design choices affect trust and rewards, rather than assuming they are neutral. The researchers put it plainly: design and presentation choices can "activate those same assumptions and biases" that shape how people treat each other, according to the research announcement.

The headline here is not that robots are sexist. It is that our habits are sticky enough to follow us into the virtual office. The good news buried in the finding is that the bias lives on our side of the screen, which means it is ours to fix. As female AI agents move from lab experiments to everyday coworkers, the pay gap they face will be a measure of us, not of them. For more on how AI is reshaping work, see GenZ NewZ's AI news coverage.