Measuring Compliance of AI-Generated Content Platforms with Data Protection Regulations: A Computational Framework and Empirical Analysis of Privacy Policy Disclosures

Abstract

The rapid expansion of AI-generated content (AIGC) platforms, encompassing large language models, text-to-image generators, and code assistants, has intensified concerns regarding whether their privacy disclosures satisfy major data protection regimes, including the EU General Data Protection Regulation (GDPR), China's Personal Information Protection Law (PIPL), and California's Consumer Privacy Act (CCPA). Despite these concerns, there remains no systematic and reproducible approach for determining the completeness and accuracy of platform privacy policies against statutory disclosure requirements. To address this gap, this paper introduces PolicyLens, a computational framework for compliance assessment that: (1) develops a machine-readable regulatory ontology representing 127 distinct disclosure obligations derived from GDPR, PIPL, and CCPA provisions; (2) uses an LLM-based information extraction pipeline to transform privacy policy content into structured evidence of compliance; and (3) aligns the extracted evidence with the ontology to generate multi-dimensional compliance scores. PolicyLens is evaluated using AIGC-Privacy-200, a newly compiled dataset containing the complete privacy policies and terms of service of 200 major AIGC platforms collected between January and March 2026. The dataset covers services headquartered across the US, EU, China, and other jurisdictions. The results show that: (i) only 23% of the platforms fully satisfy the disclosure requirements of any one regulation; (ii) obligations concerning training data provenance and opt-out mechanisms are disclosed least frequently, with compliance rates of 11% and 17%, respectively; and (iii) compared with EU-based platforms, Chinese platforms demonstrate significantly greater compliance with PIPL but lower compliance with GDPR, indicating that privacy policies tend to be drafted around the regulatory requirements of their home jurisdictions. Validation against expert legal audits conducted on a 40-platform subset shows that PolicyLens achieves an F1-score of 0.89 for obligation-level compliance detection. To facilitate further research in regulatory technology, the framework, dataset, and regulatory ontology are released as open-source resources.