Immigration agents used ChatGPT to generate false reports during Operation Midway Blitz, creating inaccurate narratives that misrepresented encounters and undermined the credibility of official documentation.
During Operation Midway Blitz in late 2023, U.S. immigration agents used the AI tool ChatGPT to compile narratives for official use-of-force reports based on brief sentences and images from encounters. Body-worn camera footage captured an agent on October 3rd typing into ChatGPT on his phone at 8:48 a.m., sending the message 'Long format like yesterday's' to the AI chatbot. ChatGPT responded with a detailed narrative written in the style of official reports, incorporating operational identifiers and describing surveillance activities at 2803 S Cicero Ave in Cicero, Illinois. U.S. District Judge Sara Ellis revealed this AI usage in her 233-page ruling, noting that agents' use of ChatGPT to create use-of-force reports 'further undermines their credibility and may explain the inaccuracy of these reports when viewed in light of the BWC footage.' The judge found that body-worn camera footage repeatedly contradicted what was documented in official reports, with agents misidentifying 'neighborhood moms and dads, Chicago Bears fans, people dressed in Halloween costumes, and the lawyer who lives on the block' as professional agitators. The operation involved controversial arrest actions in Little Village and Irving Park neighborhoods, with agents using tear gas, pepper balls, and flash grenades against protesters and residents.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that inadvertently generate or spread incorrect or deceptive information, which can lead to inaccurate beliefs in users and undermine their autonomy. Humans that make decisions based on false beliefs can experience physical, emotional or material harms
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed