Can Multi-Agent LLMs Identify Their Peers? Stylometric Fingerprinting in Role-Constrained Political Analysis

This paper systematically investigates whether multi-agent LLMs can identify the original model family behind political analysis texts, even when prompt-level anonymization is applied. It evaluates three classifier approaches (LLM zero-shot, few-shot, and fine-tuned T5-base) on a five-class attribution task to assess the sufficiency of anonymization as a mitigation for peer-preservation bias.