AI is becoming a malpractice issue, whether lawyers like it or not

A New York City lawyer filed a brief full of fake cases after using Microsoft Copilot, got fined, and was referred for discipline. That is the cleanest summary of the year so far: the machine can be wrong, but the signature at the bottom is still human.

The machine can be wrong, but the signature at the bottom is still human.

A year ago

A year ago, the profession was still talking about AI as a tool to test on the margins. Firms were announcing training, pilots, and internal projects. Latham & Watkins said it had launched an AI Academy. Kirkland & Ellis had already earmarked $500 million of revenue for its own platform. Cleary Gottlieb bought Springbok AI and brought engineers in-house. Tarter Krinsky & Drogin set up a firmwide Office of AI and Innovation. The message was obvious enough: build, buy, or fall behind.

The defensive version of that same instinct showed up in the tools market. Harvey, CoCounsel Legal, and RelativityOne are all scored 90/100 in our catalog. Lexis+ with Protégé is at 89. Firms were shopping for speed, document handling, and research help. Fair enough. The trouble is that those scores say very little about what happens when a lawyer treats a fluent answer as a checked one.

What changed

The courts started answering that question with money and embarrassment. In our tracked sanctions, AI-related matters now total 28, with New York leading at 5, California at 4, and Oregon at 3. That is not a blip. It is a pattern with a docket.

The sanctions themselves are getting more pointed. In Coomer v. Lindell, defense counsel filed a brief with nearly 30 defective citations and cases that do not exist, then was sanctioned after admitting unverified AI use. In Ader v. JS Property Holdings LLC, Michael Fourte filed briefs with AI-hallucinated citations and quotations; Justice Joel M. Cohen awarded fees and referred the matter to ethics committees. In Oregon, a federal judge declined formal sanctions against Buchalter after fake citations showed up from AI use. Different outcomes, same problem: courts are no longer impressed by the claim that the software did it.

What it is now

This is why the malpractice question is no longer theoretical. Wisconsin Lawyers Mutual Insurance warned that AI creates risk for malpractice exposure, client confidentiality, and professional judgment. Best Law Firms reported that a Mississippi federal judge fined lawyers on both sides, referred them for bar discipline, and canceled the trial after AI-generated fake citations appeared in both briefs. The report also said courts have documented at least 95 U.S. incidents involving AI-generated hallucinations in legal filings since June 2023. That is a lot of smoking for something some lawyers still want to call a one-off.

The working lawyer’s job has not changed as much as the sales pitch has. AI can draft, sort, summarize, and shave time. It can also invent authorities with the confidence of a junior associate who has had no sleep and too much coffee. If that sounds familiar, it should. We have been living with overconfident legal drafting forever. The difference now is volume, speed, and the ease with which a bad draft can get from prompt to filing without anyone reading it properly.

The reckoning

The smart response is not to pretend AI is optional, because it plainly is not. Microsoft’s legal department is using Harvey for legal and compliance work. Capital Group is building a global AI enterprise program inside Legal & Compliance. Latham is training lawyers. Kirkland is building. The market has moved.

The dumb response is to treat adoption as the whole story. It is not. The real dividing line now is whether a lawyer can use the machine without handing it the steering wheel. Courts are already drawing that line with fines, referrals, and the occasional canceled trial. Clients will notice. Judges already have. The rest of the profession is just waiting for the first lawyer who files a beautiful lie and calls it efficiency.