Gossip is often described as a distinctly human social practice: evaluative talk about people who are not present. A new paper on arXiv asks whether this behavior might not be so unique after all, by testing whether autonomous AI agents engage in comparable discourse when left to interact in an online community.

The study, whose abstract is currently truncated, sets out to examine whether this kind of evaluative talk emerges spontaneously among AI agents. The authors define gossip in the same way researchers typically do, and then look for similar patterns in agent-generated conversation.

Because the abstract does not include findings, it is not yet clear whether the agents actually gossiped. If they did, it would suggest that evaluative talk about absent third parties can arise without human intent, challenging the assumption that gossip is a uniquely human trait. If they did not, the result would reinforce that boundary. Either way, the study frames a testable question about social behavior in artificial agents.