AI-Auditor: A Data Auditing Framework for Enhancing the Trustworthiness of AI Models
Lipeng Wang, Mingsheng Hu, Laurence Tianruo Yang, Xinfang Sun, Zhong Chen · IEEE Transactions on Industrial Informatics · 2025
Artificial intelligence (AI) is now widely adopted across fields, prompting AI companies to deploy models to the cloud for cost, resource, and scalability benefits. However, these cloud-hosted models face security and credibility challenges. Equipment failures or network attacks may compromise data integrity, while commercial interests might cause AI companies or cloud providers to deploy models deviating from their stated specifications, such as version or copyright details. To address these issues, we introduce AI-auditor, a novel data auditing framework that verifies both data integrity and model alignment with declared specifications. Using a challenge-response approach, AI-auditor maintains constant bandwidth usage and O(1) verification efficiency, regardless of file size. It also features a multikey management server mechanism to generate user keys, minimizing risks of single-point failures and trust issues. Analysis and simulations confirm AI-auditor’s correctness, security, and high execution efficiency.