S.D.N.Y. sanctions attorneys for using fake ChatGPT citations
Attorneys in the Mata v. Avianca case submitted legal briefs containing non-existent case citations generated by ChatGPT. The court issued a $5,000 sanction against the lawyers for their failure to verify the AI-generated content.
Attorneys relied on ChatGPT-generated citations that did not exist.
Key facts
- What
- Attorneys in the Mata v.
- Incident date
- Jun 22, 2023
- Who
- LewKis Law
- Failure mode
- Hallucination
- AI surface
- Chatbot
- Severity
- Medium
What happened
Attorneys representing the plaintiff in Mata v. Avianca used ChatGPT to perform legal research and draft a brief. The AI generated fictitious judicial citations, which the attorneys included in their filing without verification. Upon discovering the fabrications, the court sanctioned the lawyers for their lack of professional diligence.
What broke inside the model
- 01 · TriggerA user asks for a fact, a citation, or a figure.
- 02 · Model stepThe model writes a fluent, confident answer.
- 03 · Control gapNothing ties the claim back to a real source.
- 04 · FailureA fabricated fact ships as if it were verified.
- 05 · ConsequenceThe false claim reaches a customer, a court, or the public.
Confidence holds, and even spikes, as the claim detaches from any source.
The LLM failed by hallucinating plausible but non-existent case names and legal citations. This happened because the model prioritizes probabilistic linguistic patterns over factual accuracy when generating specific references.
What it cost
Sources
- Court FilingMata v. Avianca, Inc., No. 1:2022cv01461 - Document 54law.justia.com
- Reader-SubmittedMata v. Avianca, Inc. - Wikipediaen.wikipedia.org
Cite this entry
https://failureindex.ai/failures/sanctions-attorneys-using-fake-chatgpt-citationsAI Failure Index. "S.D.N.Y. sanctions attorneys for using fake ChatGPT citations" (FI-0192). Realm Labs. https://failureindex.ai/failures/sanctions-attorneys-using-fake-chatgpt-citations (indexed Jun 5, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0192. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm would have caught this
- Prism
- OmniGuard
- AI Detection & Response (AIDR)
A runtime layer that watches the model's internal state can flag the moment a model commits to a claim it has no support for, and hold or reroute the response before it reaches a user. Realm reads those signals in real time rather than grading the transcript after the fact.