Withers v. City of Aberdeen is the rare AI-lawyering story where the court did not have to pick a side. Both sides brought their own pile of hallucinated legal authority to the party, which is convenient in the same way a kitchen fire is convenient because now everyone knows where the smoke detector is.

The case was a federal contract dispute in the Northern District of Mississippi. Tom Withers III sued the City of Aberdeen over legal fees. Each side had an out-of-state lawyer admitted pro hac vice and a Mississippi local lawyer attached to the case. That local-counsel setup is routine. It lets out-of-state lawyers appear while keeping someone licensed in the forum responsible for filings and court practice.

Then the briefs arrived.

The court found that three filings contained citations to cases it could not locate. The City's summary-judgment memorandum cited one nonexistent Mississippi case. Another City filing cited three nonexistent federal cases. Withers' opposition brief cited two nonexistent Mississippi cases. The court issued an order to show cause in December 2025 and required the lawyers to explain themselves.

By the time the sanctions order came out on June 8, 2026, the explanation was grimly familiar. The two out-of-state lawyers admitted that AI had been involved. One used an AI research tool. The other used generative AI to draft a filing. Neither verified the legal authorities before filing. The two local lawyers said they were unaware their co-counsel used AI, but admitted they had failed to review the citations in documents that bore their signatures or had been filed under their authority.

The local counsel problem

That local-counsel angle is the useful part of the story. Courts have already said many times that the lawyer who uses AI must verify the result. This order adds the less glamorous corollary: the lawyer who signs or sponsors the filing cannot treat themselves like a decorative hood ornament.

Local counsel in federal court is not just an address label with a bar number. Judge Aycock quoted the district's local rule: no eligible non-resident attorney may appear pro hac vice unless a resident attorney has been associated, and the resident attorney remains responsible to the client and for the conduct of the proceeding before the court.

That responsibility does not vanish because the resident attorney did not personally type the hallucinated cite. If your signature is on the filing, or if your practice lets someone else attach your signature without your review, the court is going to assume you meant to stand behind it. This is not some exotic AI-specific doctrine. It is basic Rule 11 dressed in modern clown shoes.

One local lawyer acknowledged that she did not check the cases after authorizing her signature on a filing. The other acknowledged that he had previously given co-counsel permission to affix his signature without reading documents, and that he should have reviewed the filings after receiving the electronic filing notice. Neither local lawyer was found to have used AI or acted in bad faith. The court still found negligence and carelessness, and that was enough for sanctions.

Sanctions with teeth

The court did not merely wag a finger and ask everyone to attend a webinar called "Please Stop Making Up Law." The sanctions changed the case.

The two out-of-state lawyers had their pro hac vice admissions revoked. They were barred from appearing in any case in the Northern District of Mississippi for two years. One was fined $2,500 and ordered to complete an AI ethics CLE. The other was fined $3,500. The two local lawyers were disqualified from further participation in the case and fined $1,000 each. The clerk was directed to send the order to the Mississippi, Louisiana, and Texas bars, and to transmit it to other courts where related attorney conduct might matter.

The court also kept the case stayed for 60 days so the parties could find new lawyers. The order directed the clerk to terminate all four attorneys as counsel of record. For a client, that is not a minor proofreading correction. That is your case getting put on blocks because every lawyer in the caption failed the same verification test in a slightly different flavor.

The trial had already been cancelled after the show-cause hearing. According to the order, it had been scheduled for March 23, 2026. The lawyers did not just embarrass themselves. They consumed court time, opposing-party resources, and client money, then forced a reset of representation in an ongoing federal case.

Why "I did not know AI could hallucinate" is dead

One lawyer reportedly told the court she was unaware AI could produce hallucinated cases and did not know what a hallucinated case was. Judge Aycock found that explanation insufficient.

That is the only reasonable answer in 2026. The legal profession has been eating this rake since Mata v. Avianca in 2023. Since then, courts have issued sanctions, standing orders, published warnings, disciplinary referrals, and enough irritated opinions to wallpaper a modest courthouse. Lawyers no longer get to act surprised that a text generator can fabricate a case citation with the serene confidence of a bad intern who found coffee.

The deeper problem is not that the lawyers used AI. The order does not announce some blanket prohibition on legal technology. It says the oldest obligation still applies: if you file a legal proposition with a citation attached, you must verify that the authority exists and supports the proposition.

AI changes the speed at which bad citations can be produced. It does not change who owns them. Rule 11 does not care whether a fake case came from ChatGPT, a legal research assistant, a half-remembered law school outline, or a dream after too much cold pizza. The signature on the filing belongs to the lawyer.

The adversarial system cannot be the QA department

Courts often discover hallucinated citations because opposing counsel or law clerks go hunting. That is a terrible QA model. It means the judicial system pays the cleanup cost after the unreliable material has already entered the record.

Withers is worse because both sides had polluted the record. There was no clean party pointing at the bad party. Everyone had homework on fire. One side's bad citations did not excuse the other side's bad citations, and one side's failure to check did not make the other side's failure less sanctionable.

This is where the "rubber-stamp" language bites. Local counsel cannot assume out-of-state counsel checked the citations. Out-of-state counsel cannot assume the AI got them right. Clients cannot assume a filing is reliable because a licensed lawyer filed it. The chain of trust has to include actual verification, not just professional letterhead passed around like a hall monitor badge.

A sane workflow is boring: mark every AI-touched legal assertion as unverified, open every cited authority, confirm the quote, confirm the proposition, confirm the jurisdiction, and confirm the case exists in the first place. If that sounds tedious, congratulations, you have rediscovered legal research.

The Withers order is memorable because it turned a bilateral hallucination into a full procedural wreck. Both litigants had to find new lawyers. The court lost time. The lawyers paid money and earned bar referrals. And a federal judge wrote another public reminder that "the AI did it" is not a defense. It is a confession that the adult supervision step went missing.