Smart-Contract-Based Access Control for AI Models in Cloud Environment

Abstract

Cloud-delivered artificial intelligence (AI) is a technology that raises many problems related to access control, such as API misuse and difficulties in managing interactions with other AI models. Centralized access control offers insufficient oversight, is prone to major administrative or system failures, and has inherent weaknesses, for example central IAM dependencies and a single point of failure. We propose a decentralized access control system that leverages blockchain and smart contracts to automatically execute agreements between parties to authenticate and maintain a history of access to the attributes of the deployed AI models. Our architecture enables access control through immutable systems, which promises that the rules and access control records can never be modified. We used an experimental methodology to preserve a real system close to production and test its performance and behavior under high load. This methodology focuses on cross-chain access control security through RoleBased Access Control that will be used in our smart contracts. From the results obtained in the previous test phase, when load was increased and RBAC was implemented in the smart contracts, we were able to measure the trade-offs between gas cost and access latency within the computational resource. Resource consumption and architecture stability were measured in milliseconds as the number of processed transactions increased. Our architecture is still applicable to cloud-based AI services. Although we gained a high gas cost from the test results due to integrating RBAC into the Ethereum smart contract, we proved that access control is guaranteed in the cloud.