1 min read · July 10, 2026
New Power Labs
As AI systems are increasingly given names, voices, and gender labels to make them feel more relatable, a 2025 study by Bazazi, Karpus, and Yasseri raises a question: What happens when the biases we hold about women follow us into those interactions?
Researchers used a version of the Prisoner's Dilemma to test how people behaved when their AI or human partner was labelled female, male, nonbinary, or no gender at all. In this game, two players choose whether to cooperate or act independently. If they cooperate, both get the best outcome. But if one chooses to cooperate and the other does not, the player who did not cooperate scores better, offering an incentive for one to “exploit” the other. If they both choose not to cooperate, both players score low.
Participants were more likely to cooperate with female-labelled partners because they expected them to reciprocate. But that same expectation also made female-labelled partners more likely to be exploited. In other words, trust became a vulnerability.
People were roughly 10% more likely to exploit an AI partner than a human one, and that gap widened when the AI was identified as female. Men were more likely to exploit across the board. Women showed stronger cooperation with female-identified partners, a pattern researchers described as homophily, or preference for those perceived as similar.
As AI begins helping founders and leaders raise capital, negotiate contracts, answer client inquiries, and support investment decisions, gendered interactions may no longer be limited to people. The next frontier of bias is the digital representatives we build and the stereotypes we unknowingly embed into every interaction.
Narinder
New Power Labs
Like what you’re reading? Subscribe to get weekly Equity Shots in your inbox.