A federal judge has revealed that an Immigration and Customs Enforcement agent used ChatGPT to draft a use-of-force narrative, then submitted the chatbot’s text as the official record of a violent encounter during their so-called “Operation Midway Blitz” raids.
The disclosure appears in a 223-page opinion from U.S. District Judge Sara Ellis, who rebuked the Department of Homeland Security for its conduct during the Chicago raids, which resulted in more than 3,300 arrests and over 600 people held in ICE custody. Ellis found repeated clashes with protesters and residents, and said agency reports frequently conflicted with what was visible on body-camera footage, rendering the written accounts “unreliable.”
Ellis wrote in a footnote that body-camera footage captured an agent “ask[ing] ChatGPT to compile a narrative” from nothing more than a brief sentence about the encounter and several images, then pasting the chatbot’s output directly into the official use-of-force report. Given the scant information provided, she noted, the AI likely filled in gaps with invented details.
Ellis noted that “to the extent agents use ChatGPT to create their use-of-force reports, this further undermines their credibility,” adding that it may account for the discrepancies seen when compared with body-camera video.
It remains unclear whether DHS policy formally restricts officers from using commercial AI tools for report writing. The agency has tested its own internal chatbot for routine administrative tasks after experimenting with systems like ChatGPT, but Ellis’s ruling indicates the officer relied on the public model—not an approved DHS tool.
Experts told the Associated Press the incident represents a “worst-case scenario” for law-enforcement use of generative AI, especially in documents meant to justify force against civilians.