Judge bars both attorneys for 2 years after hallucinated AI citations in Aberdeen dispute

A judge threw up their hands, then punished almost everybody in the room: Kathleen M. Wilson and Kathryn Y. Williams were barred for two years after they and their local-counsel counterparts submitted AI-generated legal citations that turned out to be entirely made up. The case was a dispute between Tom Withers and the city of Aberdeen, and according to the court record highlighted by lawyer Rob Freund on X, it effectively became a “comedy of AI errors” because “the attorneys admitted that the hallucinatory citations cited by them, and identified by the Court, resulted from unverified AI use.”

The immediate issue was not “AI made a mistake.” It was that both out-of-state attorneys, Wilson and Williams, used AI for research and drafting but did not verify the legal authority the tool output before filing. Williams admitted to using an AI tool to conduct legal research. Wilson admitted to using generative AI to draft her filing. The court also found gaps in the local counsel’s review process. Both Shauncey Hunter Ridgeway and Mark C. McClinton, the local counsel on the matter, admitted to “failing to review the subject filings,” and they were unable to catch the fabricated citations before they went into the court record.

To understand why a judge went so hard, it helps to look at what lawyers are supposed to do in the first place. Court filings are not just narrative. They are anchored in authority. If the citations do not exist, or if they cite the wrong propositions, the system loses its ability to test arguments. In this case, the court treated the hallucinations as an integrity problem, and the show-cause hearing functioned like the chance for an explanation and a reset. The attorneys expressed embarrassment and apologized to the Court, and they provided explanations about their independent roles in conducting legal research and/or drafting the filings at issue.

But the judge’s response was swift and concrete. Wilson and Williams, the two out-of-state attorneys, were ordered to pay fines and were barred from practicing in the district for two years. The fines were $2,500 for Wilson and $3,500 for Williams. Meanwhile, Ridgeway and McClinton were not let off lightly either: each was ordered to pay $1,000. Four lawyers, four different roles, same outcome. The court essentially concluded that the collective workflow failed in multiple places, not just in the AI stage.

And then came the broader procedural consequence: the entire matter was scrapped and the parties were given 60 days to find new counsel. Withers and the city were both given that window to replace representation, a reminder that this kind of sanction is not merely about personal discipline. It also forces a clean legal restart. In practice, that means delay risk for the client and a reputational hit for the team associated with the case, while the court preserves the credibility of the record by resetting who is responsible for what gets filed.

Why is this a boardroom story, not just a legal-curiosity story? Because the same failure pattern exists across organizations adopting generative AI. The incentive is familiar: speed up research, draft faster, reduce manual steps, and trust the output. The court record here directly undercuts that instinct. The key phrase is “unverified AI use.” If you use AI to generate citations, you still have to verify what it produced. If you do not, you do not just create a risk. You can trigger system-level consequences that include sanctions and disruption.

Second-order effects also matter. When regulators and courts increase the cost of errors, adoption patterns shift. Teams tend to add verification steps, document tooling, and tighten review checklists. That can reduce the “scramble” benefits many people chase, but it also creates defensible processes for teams that must operate under compliance expectations. For executives, the lesson is not “never use AI.” It is that AI outputs do not automatically count as validated work product, and oversight has to be designed as part of the workflow, not bolted on after the filing.

Finally, look at the roster of people involved. Williams and Wilson were responsible for drafting and assumed responsibility for the filings on behalf of their clients. Ridgeway and McClinton were responsible as local counsel and admitted they did not review the subject filings closely enough to catch the hallucinated citations. That split responsibility and shared failure should worry any organization with external professionals, in-house counsel, or vendor relationships. When the integrity of documents is on the line, every link in the chain matters. In this case, the judge turned that reality into dollars, bars, and a 60-day reset, and the warning lands hard for anyone tempted to treat generative AI like a shortcut to legal certainty.