> ## Content Index
> Fetch the complete content index at: https://www.realaigovernance.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Inside the Case That Made "AI Hallucination" a Household Legal Term
- URL: https://www.realaigovernance.com/inside-the-case-that-made-ai-hallucination-a-household-legal-term/
- Published: 2026-08-06T13:00:11.000Z
- Updated: 2026-08-06T13:00:10.000Z
- Description: AI output is a first draft from a very confident junior associate who has never read a case in their life and every governance framework should treat it that way.
- Author: Robert T. Boyer Ph.D.
- Tags: AI risk, Story - Case Study, AI Governance, Governance Debt Monthly

A federal judge asked two lawyers to produce the cases they'd cited in a motion. Neither one could.  
  
Not because the cases were hard to find. Because they didn't exist.  
  
In \*Mata v. Avianca\*, a routine personal injury dispute over an in-flight cart injury, the plaintiff's attorneys used ChatGPT to help research a legal brief. The tool returned six case citations, complete with quotations and internal reasoning that sounded exactly like real judicial opinions. The attorneys — Steven Schwartz and Peter LoDuca of Levidow, Levidow & Oberman — filed the brief without independently verifying a single one.  
  
Opposing counsel couldn't locate the cases. The court couldn't either. When Judge P. Kevin Castel ordered the attorneys to produce copies of the decisions, Schwartz turned back to the same tool that had invented them in the first place — and asked it to confirm they were real. It did. Confidently. Incorrectly.  
  
What made this case the reference point it became wasn't the hallucination itself. It was what the court found afterward: not simple carelessness, but "acts of conscious avoidance and false and misleading statements to the Court." The attorneys were sanctioned $5,000, required to notify every real judge falsely named as an author of the fake opinions, and forced to explain themselves in a public opinion that has since been cited in bar guidance, CLEs, and court orders worldwide.  
  
For firm leaders and boards, the uncomfortable part isn't that an AI tool produced false information. Tools fail. It's that the professional safeguard designed to catch exactly this kind of failure — reading and verifying your own citations before an officer of the court relies on them — didn't happen. Rule 11 doesn't have an AI exception. Neither does a partner's duty of supervision, or a firm's professional liability policy.  
  
The common objection here is, "We don't use consumer chatbots for legal research, so this doesn't apply to us." That misses the point. The failure in \*Mata\* wasn't the choice of tool. It was the absence of a verification step that would have caught the problem regardless of which tool produced it. Any firm using AI for drafting, research, or summarization without a defined check on outputs is one filing away from its own version of this case.  
  
The lesson isn't "don't use AI." It's that AI output is a first draft from a very confident junior associate who has never read a case in their life — and every governance framework should treat it that way.  
  
If your firm doesn't have a documented verification protocol for AI-assisted work product, that's the gap to close before your next filing, not after.