Preprint proposes peer-modeling controls for 6G network agents

A preprint proposes a theory-of-mind framework for LLM agents managing 6G radio-access networks, arguing that agent messages should be treated as traces of a sender's reasoning rather than objective facts. The authors model these interactions as cognitive channels and set out five design principles, including a two-level limit on peer modeling. In a signaling-storm study using locally deployed 1B-parameter telecom language models, they report that their cognitive-SNR method identified a hallucinating peer and that two-level modeling recovered the correct action.