Towards Prompt Chain Deployment in Zero Trust-enabled Compute First Networks

• Published in: IEEE transactions on consumer electronics p. 1
• Main Authors: Li, Ying, Zheng, Danyang, Fang, He, Xing, Huanlai, Chen, Xiangyi, Cao, Xiaojun
• Format: Journal Article
• Language: English
• Published: IEEE 2025
• Subjects: AIGC service deployment; Artificial intelligence; Benchmark testing; Compute first networks; Consumer electronics; cost optimization; Costs; Optimization; prompt chain; Security; Servers; Simulation; Training; Zero Trust; zero trust network
• ISSN: 0098-3063; 1558-4127
• DOI: 10.1109/TCE.2025.3582399

Compute first networks (CFN) prioritize computing resources in network planning for higher utilization and performance, making them an ideal platform for AI-generated content (AIGC) services. In AIGC applications, multiple interconnected prompts are organized into a prompt chain to complete complex tasks. Due to capacity and licensing constraints, more than one servers are required to cooperate in hosting a prompt chain. Consequently, a compromised prompt engineering-enabled server (PES) poses a significant risk to the entire CFN, potentially leading to data breaches, reduced user trust, and financial losses. Zero Trust (ZT) security architecture requires the traffic flows being validated and can efficiently address this security issue. This work investigates the problem of deploying prompt chain in ZT-enabled CFNs. To begin, we define the problem of prompt chain deployment in ZT-enabled CFNs (PCD-ZT) aiming to minimize the total service costs including deployment costs and ZT verification costs. After carefully analyzing the problem, we propose a novel factor called the verification measure (VeM) to help select proper servers hosting the prompt chain. Building on this factor, we then design the Directed sub-prompt-chAin embeddiNg (DAN) algorithm. Extensive simulation results address that DAN significantly outperforms the state-of-the-art benchmarks regarding the cost optimization.