The Stories Are Real
By now every lawyer has heard a version of the story: a brief filed with confident citations to cases that do not exist, an embarrassed hearing, a sanctions order that travels the legal internet for a week. The stories are real, they keep happening, and they should be taken seriously. Courts have sanctioned lawyers in these cases across the country, and the pattern shows no sign of stopping — because the number of lawyers using AI keeps growing faster than the number using it carefully.
But the lesson most commentary draws from these stories — that AI is uniquely dangerous and prudent lawyers should keep their distance — is the wrong lesson. It misdiagnoses what actually failed.
What Actually Failed
In every one of these cases, the failure was not that a machine produced a false citation. Machines that generate text will sometimes generate plausible falsehoods; that is a known, documented property of the technology, as knowable as the fact that a junior associate's first draft will contain errors. The failure was that a licensed attorney signed and filed work product containing citations no one had read.
Strip the technology out of the sentence and hear how it sounds: a lawyer filed a brief citing cases they never pulled. That has been sanctionable conduct since long before anyone trained a language model. The AI didn't create the duty to verify. It created a new way to feel confident while skipping it — which is a genuinely new danger, but a familiar kind.
The Discipline, Made Explicit
Careful lawyers have always practiced verification. What AI changes is that the discipline can no longer live in instinct — it has to be built into the workflow, explicitly, where deadlines can't erode it. In my own practice that looks like a few unglamorous rules. Drafting and citation-checking are separate steps, never one. Every authority gets pulled and read before it appears in anything filed — not skimmed, read. The AI is used for what it is genuinely strong at: organizing a record, surfacing inconsistencies, building first drafts on facts I supply and law I've verified. It is never treated as a source of law.
None of this is burdensome. It is perhaps an hour of process design, done once. Measured against the alternative — a sanctions order with your name on it, and the reputational half-life those carry — it may be the best-priced insurance in the profession.
The Advantage on the Other Side
Here is what the cautionary stories never mention. Once verification is built in, the calculus flips entirely. The lawyer with a disciplined system gets everything the technology promises — the file mastered in hours instead of days, the motion drafted while competitors are still organizing exhibits — with none of the headline risk, because the risk was never the tool. It was the missing step.
The profession is sorting itself, quietly, into lawyers who dismissed AI because of these stories and lawyers who read the same stories and built the discipline. Both groups think the stories proved them right. Only one of them is correct — and the gap between them widens every month.