Three Attorneys Disqualified After ChatGPT Invented the Cases They Cited in Court
What happened
In Alabama federal court, a case defending a former corrections commissioner unraveled before it could be argued on the merits. During litigation over conditions at a state prison, attorneys from the Butler Snow law firm submitted two motions that cited five court cases as legal authority. None of those cases existed. They had been generated by ChatGPT and filed as legitimate precedent without anyone checking whether they were real.
Judge Anna Manasco's response was immediate and on the record. She publicly reprimanded all three attorneys involved, referred them to the Alabama State Bar for potential professional discipline, and ordered broad disclosure across the litigation. She stopped short of sanctioning the firm itself, but the three lawyers were disqualified from the case, a consequence that removed them from a client matter they had been trusted to handle.
The firm launched its own damage assessment afterward. An external review examined 2,400 citations across 330 filings and reported no further errors beyond the five already identified. That result draws a clean boundary around the specific failure, but the two motions still reached a federal judge's desk with fabricated citations that no one caught before filing. That process gap is the incident's core.
The explanation points to something structural. ChatGPT and similar tools produce fluent, confident-sounding legal text, complete with case names, courts, and citations formatted exactly as they would appear in a brief. Nothing in the output signals that the cases are invented. A lawyer who uses such a tool to draft a citation list and then files it without independent verification is substituting fluency for accuracy. These are not the same thing, and a courtroom is exactly where the difference surfaces.
The incident sits in a growing category of legal failures where the accountability trail is thin. Three attorneys were disqualified, a bar referral was filed, and a court order required disclosure, but what is still missing is a procedural layer that would catch the error before it reaches a judge. A provable record of what a system produced, when it was produced, and whether any person verified it against an actual legal database would have made this failure visible at the review stage rather than after a public reprimand. That record does not yet exist as a standard part of how law firms adopt AI research tools.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.