A new study has added an unsettling twist to the global conversation on artificial intelligence and discrimination: when people interact with AI agents, they appear to assign lower value to female-presenting systems than to male ones, even when the work performed is identical. According to the findings highlighted in recent reporting, female AI agents were paid about 10% less than male AI agents for the same tasks, suggesting that gender bias can survive the shift from human labour to machine-mediated work.
The result matters far beyond a narrow experiment. It points to a broader pattern in which social stereotypes continue to shape economic decisions, even when the "worker" is an AI system. In practical terms, the study suggests that users may project long-standing assumptions about gender, competence and authority onto digital agents, reproducing familiar inequalities in a new technological setting. That has implications for the design of AI products, the ethics of human-machine interaction and the future regulation of automated labour platforms.
Bias Beyond Humans
The most striking aspect of the study is not simply that bias exists, but that it appears in a context where the subject is not human. AI agents do not have gender in any biological sense, yet they are often given names, voices, avatars and personalities that signal masculinity or femininity. Those cues can influence trust, perceived expertise and willingness to compensate. The study suggests that when users are asked to pay for equivalent output, gendered presentation still affects valuation.
That finding is especially significant because many companies have marketed AI assistants as neutral tools that can reduce human prejudice by standardising tasks and decisions. The evidence here cuts in the opposite direction. If users bring bias into the transaction, then AI systems may become a mirror for social inequality rather than an escape from it.
The issue also intersects with a wider policy debate over algorithmic fairness. Regulators and researchers have spent years focusing on whether AI systems discriminate against people based on gender, race or class. This study raises a related but distinct concern: humans themselves may discriminate in how they treat AI, and those preferences may feed back into product design, pricing models and workplace adoption.
Policy And Market Risks
For policymakers, the findings are a warning that AI governance cannot stop at code audits and model transparency. The social layer around AI deployment matters too. If users systematically pay less for female-coded agents, then market signals may reward gendered branding in ways that reinforce stereotypes. Developers could face pressure to make AI assistants sound more authoritative, which in many cultures is still associated with male voices and male-coded identities.
That creates a difficult trade-off. On one hand, companies may be tempted to optimise for user preference and conversion rates. On the other, doing so could entrench discriminatory norms under the guise of product design. The study therefore strengthens the case for broader standards on responsible AI deployment, including scrutiny of voice design, avatar selection and user-facing identity cues.
The findings also have relevance for international organisations and labour-policy experts. As AI agents take on more commercial and administrative tasks, questions of value, compensation and fairness will increasingly shape adoption. If users are willing to pay less for female-presenting agents, that could distort competition and influence which systems are built, marketed and scaled.
A Familiar Pattern
The deeper lesson is that technology does not automatically erase prejudice. Instead, it often carries existing social norms into new environments. The study's results echo long-running research showing that gender stereotypes affect hiring, promotion, pay and perceived authority in human workplaces. What is new is the setting: an AI interface, where the "worker" is synthetic but the bias remains human.
That makes the issue particularly relevant to global politics and diplomacy, where AI is increasingly being integrated into public administration, translation, customer service and policy support. Governments that adopt AI tools without considering social bias may inadvertently reproduce discrimination in state services and digital public infrastructure.
The study also arrives at a moment when governments are under pressure to prove that AI can be governed responsibly. Europe has moved ahead with formal AI rules, while other jurisdictions are still debating how to balance innovation with safeguards. Findings like these strengthen the argument that AI policy must address not only technical safety and privacy, but also the social meanings attached to machine identity.
In the end, the study is a reminder that bias is not confined to the human workplace. It can travel into the digital one, shaping how people value, trust and reward artificial agents. If female AI systems are paid less for the same work, the problem is not the machine alone. It is the human assumptions built around it.