While propagation-based fake news detection improves performance by modeling how news spreads, it remains vulnerable to adversarial social manipulation, where adversaries distort propagation patterns through malicious social interactions to evade detection. Existing attacks often target only one type of social behavior with predefined rules, making them vulnerable to robust defenses. To bridge this gap, we propose a new Self-evolving Social Manipulation framework (EvoSM) that utilizes Large Language Models (LLMs) to orchestrate evolutionary social manipulation campaigns for evading fake news detection. First, we design an LLM-driven Social Manipulation Orchestrator that uses a social graph interface to perceive the propagation environment, coordinate diverse manipulation actions, and refine attack strategies in context. Second, to further strengthen the manipulation capability, we introduce Reflective Experience Memory, which distills reusable manipulation insights from previous successful and failed attack attempts through reflection. Experiments on four real-world fake news datasets demonstrate that EvoSM consistently outperforms comparison baselines in attack success rates and transferability. Further analysis shows that EvoSM benefits from coordinated multi-behavior manipulation and experience-guided strategy evolution. Our study highlights the vulnerability of existing fake news detection systems to adaptive social manipulation.

The spread of fake news on social media poses a serious threat to public trust and societal stability. While propagation-based methods improve fake news detection by modeling how information spreads, they often suffer from incomplete propagation data. Recent work leverages large language models (LLMs) to generate synthetic propagation, but typically overlooks the structural patterns of real-world discussions. In this paper, we propose a novel structure-aware synthetic propagation enhanced detection (StruSP) framework to fully capture structural dynamics from real propagation. It enables LLMs to generate realistic and structurally consistent propagation for better detection. StruSP explicitly aligns synthetic propagation with real-world propagation in both semantic and structural dimensions. Besides, we also design a new bidirectional evolutionary propagation (BEP) learning strategy to better align LLMs with structural patterns of propagation in the real world via structure-aware hybrid sampling and masked propagation modeling objective. Experiments on three public datasets demonstrate that StruSP significantly improves fake news detection performance in various practical detection scenarios. Further analysis indicates that BEP enables the LLM to generate more realistic and diverse propagation semantically and structurally.