LLMs Trust Their Own: Identity-Dependent Conformity in Multi-Agent Systems

Liron Soffer, Ravid Shwartz-Ziv, Chen Shani · 2026 · arXiv

Read the source

Agents conformed more to wrong answers from their in-group — even a minimal-group label — and resisted out-group consensus, except when the out-group was human.

Questions it bears on

• Do AIs treat other AIs as their group