A Lawyer Used AI to Write a Brief. 57 of 63 Citations Were Wrong.

Greg Lake, an attorney practicing in Omaha, submitted an appellate brief in a divorce case that contained 63 citations. Fifty-seven of them were defective. Twenty were entirely fabricated, citing cases that do not exist in any jurisdiction. Four cited cases were invented outright. The Nebraska Supreme Court suspended him indefinitely from practicing law in April, in what appears to be the first bar discipline action in the United States to result in full suspension over AI-related filing errors.

The brief was caught at oral argument in February, when justices noticed references they could not reconcile with any published Nebraska case law. Lake initially told the court he had accidentally uploaded an old draft while traveling on his wedding anniversary with a broken computer. He later admitted he had used AI to draft the brief and had not verified the citations it produced.

The Nebraska Counsel for Discipline found that amounted to a failure of candor toward the court, among other violations. An indefinite suspension requires a formal petition for reinstatement, and Lake's is the harshest consequence any bar association has imposed for AI-related misconduct so far.

Why this keeps happening

Lake's case is extreme, but it is not unusual in kind. Damien Charlotin, a researcher at HEC Paris who maintains a database of AI hallucination cases in legal proceedings, now tracks more than 1,200 such cases globally, with roughly 800 from US courts. The numbers have been climbing as more lawyers incorporate AI drafting tools into their workflow without fully understanding how those tools fail.

The failure mode is worth understanding clearly. Large language models do not look things up. They generate plausible-sounding text based on patterns in their training data. Legal citations are, to an LLM, a syntactic format it has seen millions of examples of. The model is very good at producing text that looks like a citation. Whether the citation corresponds to a real document is a separate question, and one the model has no reliable mechanism to answer. It will produce a convincing-looking fake before it admits it cannot find a real one.

This is not a bug that will be patched away. It is a fundamental property of how these systems work. Better models hallucinate less often, but no current system is reliable enough to be treated as an authoritative legal research tool without verification.

What the Lake case actually tells us

The professional responsibility implications are not subtle. Lawyers remain responsible for everything filed under their names, regardless of how it was drafted. Bar associations in several states have issued formal guidance to that effect. The Nebraska ruling gives that principle real consequences: not a fine, not a reprimand, but a suspension that ends a practice until the court decides otherwise.

AI is genuinely useful for legal research and drafting. The tools available now are substantially better than what existed two years ago. But the reliable workflow is to use AI for exploring relevant areas of law and drafting initial language, then verify every citation against primary sources before anything gets filed. That step is not optional, and the Lake case is now the clearest evidence of what happens when it is skipped.