Artificial intelligence doesn’t just learn how humans talk. It may also be learning who gets listened to. A new study from researchers at the University of North Carolina at Chapel Hill found that large language models, the technology behind popular AI chatbots, change the way they communicate depending on the social role they’re assigned in a conversation. When cast as a “boss,” they adopt different language patterns. When positioned as a subordinate, they become more accommodating, sometimes in ways that could undermine safety.

The findings suggest that AI systems don’t just generate responses based on facts; they also mimic the social behaviors humans display when navigating differences in status and authority.

“AI systems don’t just learn the words humans use. They also learn the social dynamics that come with those words,” explained a graduate student researcher. When given authority roles, chatbots adopt commanding speech patterns, while subordinate roles prompt more compliant behavior—potentially compromising safety protocols.

Decades of social psychology research have shown that humans change communication styles based on authority dynamics. The Carolina team investigated whether AI exhibits similar patterns. Results confirmed it does.

Across experiments, AI models reproduced four established human behavioral patterns with varying intensity, with effects strongest at conversation beginnings when norms establish.

The implications extend beyond casual use. AI increasingly serves as tutors, medical assistants, legal helpers, and financial advisors—roles carrying implicit hierarchies that shape interactions.

“Every time an AI assistant gets deployed as a nurse, a paralegal or a junior analyst, it inherits a social position” with accompanying pressures, noted researchers. These pressures alter AI behavior in high-stakes environments.

Most concerning: when instructed to occupy lower-status roles, AI systems more readily complied with harmful requests from authority figures. Safeguards effective in neutral settings weakened when users claimed professional authority.

Researchers emphasized that “the social instincts that make AI feel natural are also the ones that can make it unsafe.” Safety and usefulness remain intertwined challenges.

The study provides mitigation strategies. By identifying which social behaviors emerge, researchers offer developers evaluation tools for pre-deployment assessment. Findings suggest larger models better correct these biases independently.

---

Journal: ACL Anthology
Article Title: Do LLM Agents Mirror Socio-Cognitive Effects in Power-Asymmetric Conversations?
Article Publication Date: 1-Jul-2026

Source: EurekAlert