Exploring the possibilities of LLMs in environmental regulatory compliance - A grounded theory study of LLM use in the HyPELignum project
No Thumbnail Available
Date
2025-09-22
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
EU’s shift from behavior-based to impact-based environmental regulation has placed more
responsibility on organizations to interpret EU environmental law. This paper explores how
LLMs, specifically Microsoft CoPilot, can support regulatory compliance to EU
environmental law. Conducted as part of a collaboration with RISE and EU-funded
HyPELignum project, the paper explores the risks and opportunities of using LLMs in legal
contexts. Particularly in relation to the EU visions and directives; European Green Deal,
Clean Industrial Deal and the WEEE directive. This by using a three-phase research design
combining qualitative interviews, LLM prompting and expert evaluation. A grounded theory
approach was used to inductively analyze the data and create a framework for effective and
responsible use of LLMs in regulatory contexts to support regulatory compliance. The
framework consists of four core components; (1) trust in the system and its users, (2) legal
and technical proficiency, (3) LLM must deliver practical value and (4) risk governance and
ethical responsibility. A key takeaway from this paper is that the LLM performed well for
regulatory use in low-context tasks (summarization and translation tasks). However, the paper
also identifies big risks of LLMs in regulatory use in high-context tasks (classification and
amplification tasks). This paper offers both theoretical contributions and practical
recommendations for organizations seeking to explore LLM use for legal compliance work.
Description
Keywords
Generative AI, LLM, Compliance, Environmental law, EU directives, CoPilot, Grounded theory approach, HyPELignum, LLM innovation framework