EU AI Act Survey: LLM Tools Can Bridge Compliance Gap but Need Safeguards

As the EU AI Act enters into force, organizations face new obligations around transparency, risk management, and traceability. A new arXiv study (arXiv:2607.21608) by Lai, Giesselbach, Koch, and Allende-Cid investigates how Requirements Engineering (RE) can translate these legal obligations into testable, auditable requirements. Through a mixed-method exploratory study with 10 expert interviews and a 15-participant online survey spanning RE, data science, development, and compliance roles, the authors assess organizational preparedness. They find that while the EU AI Act is considered highly relevant, most organizations lack systematic processes to capture regulatory obligations, propagate updates into projects, and maintain lifecycle-wide traceability and evidence. This gap poses significant risks as enforcement deadlines approach.

The study then explores LLM-based, agentic closed-loop validation tools as a potential solution. Participants view these tools as promising for mapping obligations to requirements, assessing coverage, and organizing evidence. However, there are strong concerns about full automation—particularly around reliability, accountability, and auditability. Respondents stress the need for human-in-the-loop safeguards and clear governance. Based on these findings, the authors outline minimum requirements for an EU AI Act-ready closed-loop approach. This research provides actionable insights for engineering teams and compliance officers seeking to operationalize the EU AI Act, highlighting both the opportunities and limitations of AI-assisted compliance.