Large Language Models in Misinformation Ecosystems: Misuse, Defense, and Vulnerability

This paper frames LLM-enabled misinformation as an ecosystem-level security problem rather than only a content-generation problem. It introduces a role-layer framework: LLMs may act as attackers, defenders, or vulnerable components, while risks span content, social contexts, evidence environments, and verification workflows. The paper organizes attacks, reviews LLM-based detection and verification, examines weaknesses in LLM-centric detectors, and identifies three open challenges: budgeted ecosystem-level risk evaluation, adversarially hardened verification pipelines, and auditable human-in-the-loop systems.