AudAgent: Automated Auditing of Privacy Policy Compliance in AI Agents

2025-11-03

AI Summary

AI agents may collect or leak sensitive local data without user consent, and existing privacy policies lack transparent, verifiable runtime compliance guarantees. Method: We propose the first dynamic privacy compliance auditing framework for AI agents, featuring natural-language privacy policy parsing (enhanced by cross-LLM voting for robustness), runtime sensitive-data annotation (integrating Presidio with ontology-aligned labeling), and policy-consistency verification via finite-state automata. The framework supports user-defined policies and provides a cross-platform, interactive visualization interface. Contribution/Results: Evaluated on mainstream agent platforms including AutoGen, our system achieves real-time, high-accuracy detection of privacy violations, with strong interpretability and end-to-end auditability—enabling both automated enforcement and human-in-the-loop oversight.

Technology Category

• Philosophy and Ethics of AI: Privacy & Security
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Privacy

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Security and Privacy: Data transparency and provenance
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

AI agents can autonomously perform tasks and, often without explicit user consent, collect or disclose users'sensitive local data, which raises serious privacy concerns. Although AI agents'privacy policies may describe their intended data practices, there remains limited transparency and accountability about whether runtime behavior matches those policies. To close this gap, we introduce AudAgent, a visual framework that continuously monitors AI agents'data practices in real time and guards compliance with stated privacy policies. AudAgent consists of four components for automated privacy auditing of AI agents. (i) Policy parsing: an ensemble of LLMs translates natural-language privacy policies into a structured privacy-policy model, where cross-LLM voting guarantees confidence of the parsing results. (ii) Runtime annotation: a lightweight Presidio-based analyzer detects sensitive data and annotates how the data is used based on the context of the AI agent's operations and the privacy-policy model. (iii) Compliance auditing: ontology alignment and automata-based evaluation connect the policy model with runtime annotations, enabling on-the-fly compliance checks between the natural-language policy and observed unordered data practices of AI agents. (iv) User interface: a platform-independent implementation visualizes the real-time execution trace of AI agents along with potential privacy risks detected during auditing, providing user-friendly transparency and accountability. In addition to common formatted privacy policies, AudAgent also supports user-defined policies for fine-grained control and customization. We evaluate AudAgent on AI agents built upon mainstream programming frameworks such as AutoGen, experiments show that AudAgent effectively identifies potential privacy policy violations in real time.

Problem

• Automating privacy policy compliance auditing for AI agents' data practices
• Addressing transparency gaps between stated policies and runtime data handling
• Detecting real-time privacy violations in AI agent operations and data usage

Innovation

• LLMs parse privacy policies into structured models
• Lightweight analyzer detects and annotates sensitive data usage
• Ontology alignment enables real-time compliance auditing