Trustless Enrollment: AI-Assisted zkML-Validated NFT Issuance for Secure Identity in Zero Trust Networks

Abstract

This paper presents a novel Zero-Knowledge Machine Learning (zkML)-assisted framework for secure identity enrollment in Zero Trust Network (ZTN) architectures. The proposed system addresses the limitations of static credential-based authentication by integrating zkML-driven behavioral validation with permissioned blockchain-based token issuance. A Non-Fungible Token (NFT) is used to encapsulate a one-time enrollment token (OTT) encrypted with the public key of the requesting user. The zkML layer verifies behavioral features prior to token issuance, ensuring that only users with legitimate interaction patterns receive access credentials. A permissioned Ethereum blockchain handles NFT creation and ownership management, while the enrollment process is executed through OpenZiti APIs for secure overlay network participation. Experimental evaluation shows that the zkML-validated system achieves a 96.3% fake user block rate and $98.7 \%$ NFT precision, outperforming traditional methods by significantly reducing unauthorized access. Although the zkML approach introduces a modest increase in processing time, the enhanced accuracy and security justify the trade-off. This work demonstrates the potential of combining AI-driven inference and verifiable blockchain mechanisms to achieve scalable, privacypreserving, and behavior-aware enrollment in decentralized network environments.