A client asked if you check AI-drafted citations before you file them. You said yes. In June 2026, a federal court in Mississippi found out that answer was not entirely true for four lawyers at once, and cancelled the trial because of it. This is not an AI problem. It is a verification problem, and it now has a documented cost attached to it.
“The firms that come through this transition intact will not be the firms with the best AI tool. They will be the firms with the most boring, most consistent citation-checking habit.”
What Happened in Aberdeen, Mississippi
On 8 June 2026, Judge Sharion Aycock of the United States District Court for the Northern District of Mississippi issued a sanctions order in Withers v. City of Aberdeen, a contractual dispute over unpaid legal fees. Both legal teams had built filings on generative AI, and both filings contained hallucinations: case citations that do not exist, attributed to courts that never issued them, as first reported by 404 Media.
Kathryn Young Williams admitted to using an AI tool for legal research. Kathleen Wilson admitted to using AI to draft her filing. The other two attorneys on the case, Mark McClinton and Shauncey Hunter Ridgeway, admitted to signing legal memoranda without checking the sources inside them. Aycock disqualified all four attorneys from the case, barred two of them from appearing before Northern District of Mississippi courts for two years, and imposed a combined $8,000 in monetary sanctions, per Mississippi Today. The trial did not happen. Neither side has counsel.
The Discipline Costs Nothing
You do not need a compliance department to avoid that outcome. You need one habit: every citation gets checked against a primary source before the document leaves the building, not after. That is the entire policy. It fits in a sentence, and the firm across town can already say it without checking with anyone.
A One-Page Verification Checklist for Practices Without a Compliance Team
Solo and small practitioners do not need a compliance framework to close this gap. Five checks, applied before a filing leaves the building, cover the exposure that four attorneys in Mississippi did not.
- Open every citation. Not the AI's summary of it, the primary source itself: the court website, the official reporter, or a licensed legal database.
- Treat AI-assisted drafts as unverified by default. A draft is not filing-ready until every citation has cleared the first check.
- Name one person who signs off. It does not need to be a partner. It needs to be one person, every time, so the check is never assumed to be someone else's job.
- Set a hard rule on the ten-minute citation. If a case cannot be located in a primary source within ten minutes, it does not go in the filing until it can.
- Log which AI tool produced the first draft. Not for compliance theatre. For your own pattern recognition the next time something slips through.
South Africa Has Already Decided How It Will Treat This
This is not a problem South African practitioners get to watch happen somewhere else. In Mavundla v MEC: Department of Co-Operative Government and Traditional Affairs KwaZulu-Natal [2025] ZAKZPHC 2, and again in Northbound Processing (Pty) Ltd v The South African Diamond and Precious Metals Regulator (Gauteng High Court, Case No. 2025-072038), South African courts confirmed that fabricated AI citations in court papers trigger mandatory referral to the Legal Practice Council, a pattern set out in detail by Cliffe Dekker Hofmeyr. Good faith is not a defence. Neither is time pressure. The professional duty to verify sits with the practitioner regardless of which tool produced the first draft.
What the Verify-First Model Actually Looks Like
Australia is not waiting to find out the hard way either. Habeas, a Sydney-based legal research platform, built its product around verifying against primary legislation and case law before it generates a sentence, not after. Every assertion the platform makes traces back to a citation a lawyer can open with one click. You do not need their software specifically. You need their discipline, and the adoption curve in Australia (see our briefing on the National AI Plan) suggests South African firms have somewhere between 12 and 18 months before that discipline stops being a differentiator and starts being the baseline every client assumes you already have.
Holkam Advisory tracks the regulatory and disciplinary signals shaping how AI is used in legal practice across Australia and South Africa. If you want to know where the verification bar is likely to sit in twelve months, not where it sits today, contact us here.