A Multi-Agent Ai Framework For Detecting Coordinated FraudCampaigns In Digital Media

Abstract

Coordinated fraud campaigns in digital media spanning ad-fraud (clicks, impressions, installs), coordinated
inauthentic behavior (CIB), astroturfing, and manipulation networks—have evolved into adaptive, multistage
operations that exploit platform affordances, automation, and cross-channel amplification. Singlemodel
detectors (e.g., isolated bot classifiers or per-event anomaly scoring) are often brittle against
adversarial drift, data sparsity, and the inherently relational nature of coordination. This review synthesizes
the state of the art and proposes a multi-agent AI framework that operationalizes coordination detection
as a pipeline of collaborating agents: (i) ingestion and provenance agents, (ii) feature and representation
agents, (iii) coordination-graph construction agents, (iv) anomaly and campaign discovery agents, (v)
evidence fusion and adjudication agents, and (vi) monitoring and governance agents. We ground the design
in established findings on coordinated information operations, coordination archetypes and temporal
stability, and practical coordination toolkits, while incorporating advances in dynamic-graph anomaly
detection, graph neural approaches to fraud , and evaluation pitfalls in bot detection. We provide a “cameraready”
research agenda covering robust temporal/multimodal fusion, agent-to-agent calibration,
explainable coordination evidence, and responsible deployment under uncertainty.