FedAuth: Collaborative Authorization for Federated Learning Models via Secret Sharing

Abstract

The development of high-value federated models requires the collaborative efforts of numerous federated nodes, each contributing substantial computing and data resources. Therefore, it is crucial to protect the intellectual property rights of federated models with multi-party participation. Authorization verification provides a practical way to govern the use of collaboratively trained federated models and protect their intellectual property. However, existing methods mainly address single-owner authorization, where the authorization proof is usually controlled by one party and cannot be naturally partitioned into independent shares for multi-node federated scenarios. In this paper, we study the authorization problem of federated learning and propose FedAuth, a collaborative authorization scheme that enables federated nodes to jointly participate in authorization decisions. Specifically, we design an authorization proof generation mechanism for federated models, treating the parameters used to produce noise-based trigger sets as the authorization proof and adopting a block-partitioned backdoor watermarking strategy to strengthen resistance to proof forgery. To support collaborative authorization, we use secret sharing to design an authorization proof distribution and verification mechanism, in which the authorization proof is split into multiple independent shares and distributed to federated nodes. A user completes authorization only after successfully collecting a sufficient number of valid proof shares from federated nodes. FedAuth provides threshold-based authorization verification for service-mediated federated models. The service-side access-control component uses the verification result to permit or reject normal inference requests. Theoretical analysis and experiments show that FedAuth preserves model utility and maintains robust authorization signals under common model modifications.

Keywords

Federated learning / model authorization / copyright protection / secret sharing

Rights & permissions

Higher Education Press 2026

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.