For two years, stories about lawyers filing briefs with cases that do not exist have been treated as a punchline. In June a federal judge in Mississippi stopped laughing. She sanctioned every lawyer in one case, on both sides, because both sides had filed AI-generated citations that nobody checked. One of the firms involved had a written AI policy. It did not help, because nobody followed it.
What happened
Bloomberg Law reports that Senior Judge Sharion Aycock of the Northern District of Mississippi issued the order in June 2026 in a fee dispute between an attorney and the city of Aberdeen. At a hearing in January, two of the lawyers admitted using AI and said neither had verified the authorities before filing. The judge found all four lawyers violated Rule 11, the federal rule that makes a lawyer vouch for what they file. Two were barred from appearing in that district for two years and fined $2,500 and $3,500. The two lawyers who had sponsored them were removed from the case and fined $1,000 each.
Lawyer Monthly adds detail that matters for anyone running a business. One lawyer, a solo practitioner, said she did not know AI could invent citations, and had been disciplined for similar conduct by a Louisiana bankruptcy court in April. The judge called her repeat misuse an "extreme dereliction of professional responsibility." The other was a partner at a firm whose AI policy required lawyers to verify research. She did not follow it. The firm's managing partner told the publication it requires verification of everything in a filing and will keep training its team.
Mississippi is not an outlier. A Mondaq analysis published in March describes the Sixth Circuit Court of Appeals sanctioning two lawyers over briefs in three appeals that cited nonexistent cases and misstated the record. The penalties included a $45,000 payment to the court plus the other side's legal fees, and a referral for possible discipline. The author calls it likely the largest sanction yet for AI-hallucinated briefs.
Why it matters for your business
You may never file a brief. But the pattern in these cases is the same one that shows up in sales proposals, HR letters, customer emails and compliance reports.
The tool is not the problem. Skipping the check is. Both the Mississippi lawyers and the appellate lawyers had access to accurate research. They used AI output as if it were finished work. AI writing tools are built to produce text that sounds right. That is different from text that is right.
A policy on paper is not a defense. The firm with a verification rule still saw its partner sanctioned. If your AI policy lives in a handbook nobody opens, it will not protect you when an AI-written error reaches a client or a regulator.
Your name is on it. Courts keep saying the same thing: the person who signs is responsible, no matter who or what drafted it. The same is true when your company sends a contract summary, a safety notice or a quote.
If you hire lawyers, ask the question. Courts are now sanctioning lawyers on both sides of a case. It is fair to ask outside counsel how they use AI and how they check it. This is not legal advice, just a sensible vendor question.
What we'd do
- Name the high-stakes documents. Anything sent to a court, regulator, client or the public gets a human check against original sources. Write that list down.
- Make checking cheap. Choose tools that show their sources with links, and train staff to click them. A citation you cannot open is a red flag.
- Assign an owner. Each AI-assisted document should have one named person who signs off. No owner, no send.
- Spot-check, then say so. Once a month, pull a few AI-assisted documents and verify them. Share what you find. That is how a policy becomes a habit.
- Teach the failure mode. One lawyer said she did not know AI could make things up. Ten minutes of training would have saved her two years of practice in that court.
Our view
You could argue that judges are being harsher on AI mistakes than on old-fashioned sloppy research, and there is something to that. A wrong citation is a wrong citation. But the courts are right to send a loud signal, because AI makes confident errors faster and in greater volume than any tired associate. The answer is not to ban the tools. It is to treat every AI draft as a draft, put a named human between it and the outside world, and make sure that human actually reads it.
Want a system like this in your business?
A discovery conversation is where we map it: what to automate first, where the human checkpoints go, and what it would take.
Book a discovery conversation →