A Federal Judge Just Flagged ChatGPT in Immigration Use-of-Force Reports
What happened
U.S. District Judge Sara Ellis sits on the bench in Chicago, and in late November she said something that should worry every agency writing reports with a chatbot: immigration agents had been running their use-of-force paperwork through ChatGPT, and the results were riddled with errors. The setting matters. These reports were produced during an active immigration crackdown, amid public protests, at the exact moment when the public most needed to trust what officers wrote down about their own conduct.
Use-of-force reports exist for one reason: to create an honest record of what happened when an officer used physical force against a person. That record gets read by prosecutors, by defense attorneys, by oversight boards, sometimes by juries. When a language model generates the narrative, the document stops being a firsthand account and becomes a paraphrase, one that can invent details, smooth over contradictions, or misstate facts the officer never said. Judge Ellis's criticism wasn't a stylistic complaint. It went to whether the reports could be trusted at all, and whether sensitive information about people involved in enforcement actions had been fed into a commercial tool without regard for who else might see it.
The deeper problem isn't that agents used a chatbot. It's that nobody appears to have checked the output against reality before it entered a legal record. No agency policy requiring a human to verify names, dates, and use-of-force sequencing against body-camera footage or officer notes. No log showing which report was AI-drafted versus human-written. When that verification step is missing, credibility doesn't erode gradually. It collapses the moment a judge, a reporter, or a defense attorney catches the first factual error, and every prior report from that agency becomes suspect too.
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.