TOSense: We Read, You Click

Terms of Service (ToS) are often lengthy and complex in terms of terminology, and users often click ‘Agree’ without reading, which poses privacy and compliance risks. We propose TOSense, an open-source browser plug-in leveraging large language models to assist users in interactively understanding ToS content. The system adopts a dual-mode architecture, integrating MiniLM and BART models, and incorporates a lightweight ToS information acquisition script, ‘tos-crawl’, to automatically extract, clean, and structure the ToS content of web pages. We designed a data-independent risk assessment process (REP) and verified it on the ToS of five mainstream platforms. The prototype is publicly available, and we conclude by discussing directions for future work.