Programs Certification Faculty Corporate Training Research & Labs AI & Law Quiz Apply Now →
Home / Resources / Risk
Risk

AI hallucinations in filings — and how to avoid sanctions.

Lawnova Academy · Updated June 2026

In 2023, lawyers in Mata v. Avianca filed a brief citing cases that did not exist. A chatbot had invented them — complete with quotations and fake citations — and the lawyers hadn't checked. The court imposed sanctions, and the episode became a cautionary tale repeated in courtrooms worldwide. The lesson is not "never use AI." It's "never file what you haven't verified."

Why language models hallucinate citations

A large language model predicts plausible text; it does not look up authorities in a database unless it is specifically connected to one. Asked for supporting cases, a general chatbot will produce text that looks exactly like a citation — correct reporter format, realistic party names, confident holdings — because that is the pattern it learned. Plausibility is not truth. The output can be entirely fabricated and still read perfectly.

It keeps happening

Mata v. Avianca was not a one-off. Courts across multiple jurisdictions have since flagged briefs containing fictitious or misquoted authorities generated by AI, leading to sanctions, fee forfeiture, and referrals. Many judges now issue standing orders requiring disclosure of AI use and/or certification that all citations were verified.

The habit that prevents it

  1. Treat AI output as an unverified draft. Useful for structure and ideas — never authoritative on its own.
  2. Open every citation in a real source. Confirm the case exists in an official reporter or a primary database (not a screenshot, not the chatbot's own restatement).
  3. Check the holding, not just existence. A real case cited for a proposition it does not support is its own problem.
  4. Verify quotations verbatim. Hallucinated quotes appear even in real cases.
  5. Validate with a citator (KeyCite/Shepard's) for treatment and good law.
Retrieval-grounded legal tools — which search an actual case database and cite back to it — reduce hallucination risk dramatically compared with consumer chatbots. But "reduce" is not "eliminate." The lawyer's verification is the control that matters.

If you use AI, document it

Keep a simple record of which tool produced what, and that a lawyer verified each authority. If a court asks — and increasingly they do — you want a clear answer. This also satisfies the supervision and candor duties discussed in our ethics overview.

Learn the verification workflow hands-on

Level II — AI Research & Drafting teaches citation verification, retrieval-grounded research, and drafting under oversight, with faculty-led labs.

See Research & Drafting

Keep reading

Can Lawyers Use AI? The Ethics Rules, Explained
Competence, candor, supervision — the duties behind the headlines.
How to Verify AI Legal Research
The step-by-step citation-checking process, in detail.

Educational, not legal advice. This article is general information about AI in legal practice and does not constitute legal advice or create any attorney–client relationship. Consult the rules of your own jurisdiction and your professional-responsibility counsel before acting.