PrivacyLens: AI-Powered Privacy Policy Analyzer

Almost every online service asks users to accept a privacy policy, and almost nobody reads it. The documents are long, written in legal language, and spread their most important statements who receives the data, how long it is kept, how to opt out across many pages. This paper presents PrivacyLens, an AI-powered privacy policy analyzer that accepts a policy as a URL, an uploaded document or pasted text and returns a normalized 0 to 100 risk score, findings across eight privacy categories, dark-pattern indicators, an exposure ledger, a side-by-side policy comparison and a policy-aware AI assistant. Instead of trusting a single generative answer, PrivacyLens runs a staged hybrid pipeline: topic classification based on the OPP-115 taxonomy (Model A), a severity cross-check against human-reviewed ToS;DR data (Model B), negation-aware deterministic rules, and a large language model as a fallback and explainer. The system is built with React, FastAPI, MongoDB Atlas, Firebase Authentication, Playwright and Groq-hosted LLMs, and is also delivered as a Chrome extension. In preliminary testing, all four recorded analysis requests returned HTTP 200 with a median response time of 7.51 s (mean 12.01 s, range 6.66–26.35 s), and an end-to-end extension scan of a real shopping website produced an 81/100 High Risk report with concrete findings that were stored and later retrieved from History. We also provide a qualitative comparison with prior work and a plan for further evaluation.