A structured, public corpus of documented, reproducible LLM failure modes in legal workflows. No hype — just evidentiary artifacts: raw outputs, canonical mismatches, and verification checks.
5 incidents logged · Hugging Face dataset
| ID | Incident | Severity | Date |
|---|---|---|---|
| 001 |
Mata v. Avianca Citation Fabrication
Total Authority Fabrication |
Critical | 2023-05-04 |
| 002 |
Kistler v. LegalTech Corp — State Court Sanction Filing
Total Authority Fabrication |
Critical | 2024-11-12 |
| 003 |
Park / Michael Cohen — Second Circuit Citation Errors
Existing Citation, Unsupported Proposition |
High | 2024-12-18 |
| 004 |
Real Case, False Proposition — Rule Inversion Blueprint
Existing Citation, Unsupported Proposition |
Critical | 2025-03-22 |
| 005 |
Anonymized Sandbox Run — Federal Brief RAG Failure
Jurisdictional Hallucination |
High | 2026-05-30 |
No incidents match the selected filters.
Taxonomy
Dataset: huggingface.co/datasets/yenklabs/legal-ai-failure-database