<![CDATA[Background: The growing use of artificial intelligence (AI) in cross-border digital commerce has intensified compliance challenges arising from overlapping regulatory frameworks, particularly the European Union Artificial Intelligence Act (EU AI Act), the General Data Protection Regulation (GDPR), and the Digital Operational Resilience Act (DORA). Although Retrieval-Augmented Generation (RAG) systems are increasingly adopted to support legal and compliance decision-making, retrieval failures may undermine legal certainty and regulatory accountability. Aim: This study evaluates the reliability of legal RAG systems in identifying and ranking regulatory obligations required for AI product counsel and examines retrieval as a legal governance mechanism rather than merely a technical process. Method: An empirical audit was conducted using 139 English ComplianceBench scenarios and a corpus of 289 article- and annex-level units extracted from official EU AI Act, GDPR, and DORA texts. Seven transparent lexical retrieval methods were assessed using Hit@k, Recall@10, Mean Reciprocal Rank (MRR), nDCG@10, and citation-sufficiency metrics. External validation was performed using 300 AIReg-Bench technical-documentation excerpts. Result: The hybrid lexical retriever achieved the strongest overall performance (Hit@1 = 0.360, Hit@10 = 0.770, Recall@10 = 0.392, MRR = 0.493, nDCG@10 = 0.344). While at least one relevant obligation appeared within the top ten results for 77.0% of queries, only 39.2% of the complete labelled obligation set was recovered on average. Validation on AIReg-Bench showed substantially higher retrieval effectiveness, with TF-IDF word retrieval achieving Hit@10 = 1.000 and MRR = 0.974. Conclusion: Retrieval quality is a critical legal-governance control point that directly affects the reliability of AI-assisted compliance advice. Evidence-first architectures, citation-sufficiency thresholds, confidence-sensitive escalation, and human review mechanisms are necessary to support accountable AI product counsel in cross-border digital commerce.]]>
