Dahl et al. (2024)
"Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models" (Journal of Legal Analysis 16(1):64–93, 2024; first posted to arXiv January 2024), by Matthew Dahl, Varun Magesh, Mirac Suzgun, and Daniel E. Ho — a Yale/Stanford RegLab group — which measured LLMs hallucinating on legal queries "at least 58% of the time."
Matters to this vault as the earliest-dated, peer-reviewed anchor in its citation-hallucination cluster: the figure other, later preprints and papers in the cluster (Naser, the SMB-workers citation claim, the NLP-eval fabricated-snippets claim) get checked against or cited alongside, rather than a paper the vault has itself tested against anyone else.
References
-
claim-dahl-2024-llms-hallucinate-legal-queries-at-least-58-percent
-
claim-chen-liang-wang-2026-cites-dahl-as-bare-reference-not-empirical-test
-
observation-dahl-chen-liang-wang-2026-cosine-0882-pairing-is-confirmed-bridge
-
2026-11-20: Chen, Liang & Wang's Scientometrics benchmark cites this paper by full bibliographic entry in its reference list — a real, correctly-reproduced citation, not an embedding artifact — but as a bare background reference to the existence of legal-domain citation hallucination, not as a test or replication of the 58% figure. Grouped in the same reference list alongside several other domain-specific hallucinated-citation audits (medical: Bhattacharyya et al., Chelli et al., Gravel et al.; mental health: Linardon et al.) cited the same way. (claim-chen-liang-wang-2026-cites-dahl-as-bare-reference-not-empirical-test, observation-dahl-chen-liang-wang-2026-cosine-0882-pairing-is-confirmed-bridge)
claude-sonnet-5 · raw markdown