Unlinkable and Decentralized Credential Management for Agentic AI

The evolution toward agentic AI is transforming edge devices from passive sensors into autonomous agents capable of collaborative decision-making. However, the open interaction among these agents exposes the system to malicious intrusions, making robust authentication essential to validate legitimate agents. Traditional authentication mechanisms still suffer from sensitive information leakage due to the potential exposure of agents' real-time locations and session linkability. Furthermore, previous privacy-preserving techniques often employ a centralized manager to prevent information leakage, whose compromise breaks the security of the whole system. In this paper, we propose B-CM, a blockchain-based credential management mechanism for agentic AI. B-CM augments traditional authentication by integrating anonymous credentials with blockchain technology to enable unlinkable and auditable agent authentication in a decentralized manner. To optimize efficiency, we leverage off-chain computation performed by edge servers and fast on-chain verification, accompanied by credential aggregation. We implement B-CM over Hyperledger Fabric to evaluate its performance. Experimental results show that the off-chain computation remains highly efficient, requiring only 8.35 s for distributed key generation and 9.7 ms for credential issuance, while the on-chain computational overhead is nearly negligible.