Future 6G networks will rely on large language model (LLM) agents to manage the radio access network (RAN). Existing designs treat inter‑agent messages as objective facts, but each message is actually a trace of the sender's reasoning and carries a subjective conclusion. Even a syntactically valid report can spread an AI hallucination and trigger a cascade of failures invisible to protocol checks. Before acting, a receiver must possess a theory of mind (ToM) – it must infer what the peer believes and what it should have believed in that context. By modeling these interactions as cognitive channels on a cellular sheaf we obtain a unified resilient multi‑agent framework, from which five design principles emerge: (i) a message is evidence of hidden reasoning; (ii) trust is a continuous cognitive signal‑to‑noise ratio (SNR), i.e., asserted precision over deviation from the modeled peer belief; (iii) network‑wide consistency and resistance to hallucination contagion are computable via the sheaf's Laplacian; (iv) peer‑modeling should stop after exactly two levels to save compute and survive mutual‑information decay; (v) credible capacity is bounded by alignment with operational goals, not link bandwidth. A signaling‑storm study on a locally deployed 1‑billion‑parameter telecom language model validates the theory: cognitive SNR isolates a hallucinating peer that three of its four neighbors agree with; a divergence gate ranks every wrong peer above the right one; only depth‑two ToM recovers the correct action; and the spectral gap decides whether a topology reaches consistency within the near‑real‑time budget.
Review