AI in legal: the Mata lesson.

Mata v. Avianca, the 2023 federal case where an attorney filed a brief citing six fabricated court decisions generated by ChatGPT, was not a one-off mistake; it was a structural property of LLMs deployed without scope containment. Three years later, our own field reports show the same failure pattern in a Boca Raton estate planning chatbot that fabricated a 2022 Florida elective share amendment. The fix was never smarter AI. It was hard scope: every out-of-scope question routes to a consultation booking, every response carries a general-information disclaimer, and the model never attempts to answer anything it was not explicitly trained on. The longform with the deployment checklist ships in June 2026.