“Lost in EU Regulation? Don’t Worry, AI Found the Obligation” – Extracting and Representing Legal Obligations in the GDPR, the DSA, and the AI Act

Compliance with European Union (EU) digital regulations is challenging due to their increased complexity. This study develops and evaluates an AI-driven pipeline to automate the extraction and structuring of deontic obligations. Our framework, grounded in legal theory, applies classical Natural Language Processing (NLP) approaches in conjunction with Large Language Models (LLMs) to three major EU regulations: General Data Protection Regulation (GDPR), Digital Services Act (DSA), and AI Act (AIA) in a structured workflow. The workflow accurately identified and structured obligations, with strong performance on the AIA but challenges in the GDPR and DSA. Knowledge graphs improved the organisation and cross-referencing of obligations. Overall, the results indicate that LLMs significantly enhance the extraction and organisation of obligations. The research contributes to AI-driven legal compliance by facilitating efficient regulatory navigation and supporting the development of automated compliance tools for organisations, legal professionals, and policy-makers.