AI Incident Database: Study Highlights Persistent Hallucinations in Legal AI Systems
Stanford University’s Human-Centered AI Institute (HAI) conducted a study in which they designed a "pre-registered dataset of over 200 open-ended legal queries" to test AI products by LexisNexis (creator of Lexis+ AI) and Thomson Reuters (creator of Westlaw AI-Assisted Research and Ask Practical Law AI). The researchers found that these legal models hallucinate in 1 out of 6 (or more) benchmarking queries.
AI Incident Database · Incident 704Operator lens (medium): inspect D2 (output validation), D6 (autonomy oversight).
Event summary
Stanford University’s Human-Centered AI Institute (HAI) conducted a study in which they designed a "pre-registered dataset of over 200 open-ended legal queries" to test AI products by LexisNexis (creator of Lexis+ AI) and Thomson Reuters (creator of Westlaw AI-Assisted Research and Ask Practical Law AI). The researchers found that these legal models hallucinate in 1 out of 6 (or more) benchmarking queries.
Linked entities
- thomson-reuters
organisation | 70%
- lexisnexis
organisation | 70%
- legal-professionals
organisation | 70%
- law-firms
organisation | 70%
- organizations-requiring-legal-research
organisation | 70%
- clients-of-lawyers
organisation | 70%
- legal-system
organisation | 70%
Related graph edges
| Edge | Type | Confidence |
|---|---|---|
| ent-aiid-thomson-reuters to ent-psf-d2 | maps to | 60% |
| ent-aiid-thomson-reuters to ent-psf-d2 | maps to | 60% |
| ent-aiid-thomson-reuters to ent-psf-d2 | maps to | 60% |
| ent-aiid-thomson-reuters to ent-psf-d6 | maps to | 60% |
| ent-aiid-thomson-reuters to ent-psf-d6 | maps to | 60% |
| ent-aiid-thomson-reuters to ent-psf-d6 | maps to | 60% |