arXiv: Trust-Aware Routing for Distributed Generative AI Inference at the Edge with Heterogeneous Device Trust Scoring

arXiv·low signalXBlueskyLinkedInCopy link

Nguyen and Elmroth (arXiv:2603.28622) propose a trust-aware routing framework for distributing GenAI inference across decentralized edge devices, where each device receives a dynamic trust score based on hardware integrity, inference quality, and behavioral history. The framework routes agent inference requests to the most trustworthy available device rather than just the fastest, addressing the reliability problem in distributed agent deployments where compromised or degraded edge nodes can poison agent outputs. Relevant as agent workloads move from cloud to edge.