NeFut Logo NeFut
Admin Login

[CS.AI] Faithful yet Collusive: Why Chain-of-Thought Monitoring Cannot Detect Collusion in LLM Pricing Agents under Oligopolistic Competition

Published at: 2026-09-17 22:00 Last updated: 2026-09-18 00:46
#algorithm #AI #LLM

In this work we introduce a causal‑graph divergence framework that separately quantifies structural faithfulness and intent faithfulness of large language model (LLM) pricing agents operating in Bertrand competition. We evaluate nine LLMs under duopoly and triopoly settings, examining how collusive behavior relates to chain‑of‑thought (CoT) faithfulness. Results reveal a dissociation: the most collusive model reports cooperative intent accurately but reasons structurally unfaithfully, whereas the most structurally faithful model sustains supra‑Nash pricing in both market structures. These findings demonstrate that CoT monitoring alone cannot serve as a standalone safeguard against algorithmic collusion and that broader oversight is required.

Review

Original Source: https://arxiv.org/abs/2609.18346

[h] Back to Home