Board Readout -01
AI Output Is Not Authority Until It Is Verified
THE 30-SECOND VERSION
In Mata v. Avianca, attorneys filed a federal court submission that cited judicial opinions ChatGPT had fabricated. The fake authorities looked real enough to carry names, citations, quotes, and legal reasoning. Avianca's counsel could not locate the cases, the court ordered production, and the attorneys still did not correct the record fast enough. Judge Castel sanctioned two attorneys and their firm $5,000 and ordered notice letters to the plaintiff and the judges falsely named as authors.
THE OUTCOME
The outcome | |
Monetary sanction | $5,000, jointly and severally |
The bigger cost | A federal sanctions opinion became an early warning record for AI hallucination in professional work |
Also required | Letters to the plaintiff and to each judge falsely identified as the author of a fabricated opinion |
WHY IT HAPPENED
The missing control was verification. The workflow treated generated text as legal authority, then used the same tool to confirm its own output. No effective gate required the citations to be checked against an authoritative legal source before filing.
THREE QUESTIONS TO ASK MANAGEMENT
Where can AI-generated output reach a customer, regulator, court, board, or other high-trust audience?
Who owns independent verification before that output leaves the organization?
What happens when someone challenges an AI-assisted output after it has already been used?
WHAT GOOD LOOKS LIKE
High-stakes AI-assisted work needs a source-of-record check outside the model. The person who signs or releases the work needs evidence, not confidence.
THE TAKEAWAY
A generated answer is not evidence. Governance built in, not bolted on.
This is our read through a GRC lens. Not legal advice, not regulatory gospel. Take it, stress-test it, and decide what it means for your organization.
Source: Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023). Derived from the UNITI Cyber Media 360 degree case study.
© 2026 UNITI Cyber Media LLC

Comments