Smart Privacy Policy Assistant: An LLM-Powered System for Transparent and Actionable Privacy Notices

ACL ARR 2026 January Submission6533 Authors

Keywords: privacy policies, large language models, interpretable NLP, usable privacy, risk assessment, decision support

Abstract: Most users consent to online privacy policies without reading or understanding them, despite the significant implications these documents have for personal data collection, sharing, and monetization. Privacy policies are typically long, legally complex, and difficult for non-expert users to interpret, limiting their effectiveness as mechanisms for informed consent. We present the Smart Privacy Policy Assistant, an LLM-powered system that automatically analyzes privacy policies to extract and categorize key clauses, compute human-interpretable privacy risk scores, and generate concise, clause-grounded explanations. The system is designed for real-time use through browser extensions or mobile interfaces, surfacing contextual warnings when users are asked to disclose sensitive information or grant permissions. We describe the end-to-end pipeline, including policy preprocessing, schema-guided clause analysis, monotonic risk scoring, and explanation generation. Experimental results on real-world privacy policies show strong agreement with human annotations at the clause level and meaningful alignment with human judgments of policy risk, while a user study demonstrates improved comprehension and decision-making. Together, these findings suggest that structured, interpretable LLM-based analysis can make privacy disclosures more transparent and actionable.

Paper Type: Short

Research Area: Human-AI Interaction/Cooperation and Human-Centric NLP

Research Area Keywords: Privacy, Ethics, and Fairness, Natural Language Processing Applications, Interpretability and Explainability in NLP

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Data analysis

Languages Studied: English

Submission Number: 6533