CDAP-PFMs: A Cross-Domain Authentication Protocol for Edge Inference-Based PFMs

The Pre-trained Foundation Models (PFMs) based on edge inference reduce communication latency from the cloud server to the user mobile devices, and support different edge nodes to provide adaptive Artificial Intelligence (AI) services to the users. The complex network environment has attracted widespread attention to the communication security of cross-domain AI services. Authentication protocol is an important way to ensure the communication security among entities. However, existing cross-domain authentication protocols have not taken into account the communication delay caused by complex cryptographic operations on the mobile devices. To address these issues, a Cross-Domain Inference Authentication (CDIA) protocol for PFMs is proposed in this paper. The complex cryptographic operations are executed collaboratively by the servers from different domains to realizes decentralized authentication. Besides, the correctness and security of CDIA protocol is proved by formal analysis. The security of CDIA protocol is verified by using the Automated Validation of Internet Security Protocols and Applications (AVISPA) tool. The security of negotiated group key is proved by Real-or-Random (RoR) model. Finally, the performance of CDIA protocol is evaluated to show its efficiency. Compared with related protocols, the computing cost and communication overhead of CDIA protocol are at least 52.38% and 11.9%, respectively.