A Privacy-Preserving Scheme Based on Attribute and Homomorphic Encryption for Trustworthy Multi-Agent Data Management in Intelligence Edge Networking

Abstract

With the deep integration of AI-native edge intelligence and 6G networks, distributed multi-agent systems composed of autonomous agents such as digital twins, autonomous vehicles, and the industrial Internet of Things are becoming the critical infrastructure for achieving “Synesthesia of Machines” of multi-source heterogeneous data. These agents need to collaboratively integrate multi-source heterogeneous data in resource-constrained edge environments to perform real-time perception, reasoning and decision-making. However, the distributed and dynamic nature of multi-agents systems significantly expands the attack surface, exposing severe security and privacy challenges. Although blockchain technology provides anti-tampering and traceability, its openness and transparency also introduce privacy risks. We propose a trustworthy data management framework for multi-agent systems, which uses ciphertext-policy attribute-based encryption (CP-ABE) with proxy re-encryption (PRE) to build a flexible access control mechanism, enabling secure data sharing among agents while preventing sensitive information exposure. We further employ Paillier homomorphic encryption to enable verifiable data computations in the ciphertext domain, ensuring that data remains usable yet invisible, and providing cryptographic assurance for trusted collaboration among agents. In addition, the combination of IPFS and blockchain technology is used to alleviate the storage burden brought by a large amount of data. Meanwhile, an access whitelist is established to mitigate the extra resource expenditure resulting from recurrent authentication. The experimental results confirm that our approach not only fulfills the low-energy and low-latency demands of edge intelligence scenario but also provides formal guarantees for privacy and verifiability in distributed multi-agent networks.