A paper published on March 7, 2026 presents a benchmark with 4,876 criminal-law passages and 100 complex questions written by legal experts to assess retrieval in legal RAG systems.
Published on March 7, 2026, the paper introduces Legal RAG Bench, a domain-specific benchmark for assessing retrieval in retrieval-augmented generation systems focused on criminal law. Its dataset contains 4,876 passages in the field and 100 complex questions written by legal experts. The authors say retrieval problems can contribute to fabricated answers.
The work is relevant to teams evaluating legal systems: it provides a reference set for testing content retrieval, but does not by itself establish the quality of a system in use. Consult the original paper to check its full method and findings. When studying or applying its ideas with AI, avoid submitting confidential legal documents without authorization and an appropriate policy.