According to Fast Company, an immigration agent reportedly used ChatGPT to generate official law enforcement reports from just a single sentence and photographs. Ian Adams, an assistant criminology professor at the University of South Carolina, called this approach “the worst of all worlds” and a “nightmare scenario” that goes against all established AI guidance. The Department of Homeland Security hasn’t commented on whether it has AI use policies, and the body camera footage referenced hasn’t been released. Experts warn this raises serious concerns about accuracy, privacy, and professionalism in law enforcement contexts where objective documentation is critical.
Why this is terrifying
Here’s the thing: when courts evaluate whether force was justified, they rely heavily on the specific officer’s perspective in that exact moment. Adams explained that using AI to generate reports essentially “begs it to make up facts” in high-stakes situations. You’re replacing an officer’s actual observations with what an AI model thinks probably happened. And that’s a huge problem when someone’s rights or safety are on the line.
But wait, it gets worse. If the agent used the public version of ChatGPT, he probably didn’t realize he lost control of those uploaded images the moment they hit the system. Katie Kinsey from NYU’s Policing Project noted this could mean sensitive law enforcement evidence entering the public domain. Basically, you’re handing potential evidence to bad actors while compromising investigations.
The professionalism question
Andrew Guthrie Ferguson, a law professor at George Washington University, raises another critical point: “Are we OK with police officers using predictive analytics?” The concern is that AI models generate what they think should have happened, not necessarily what actually occurred. And when those AI-generated narratives end up in court justifying police actions, we’ve crossed into dangerous territory.
Look, even tech companies that sell AI to police departments recognize the limitations. Companies like Axon specifically avoid using visual components in their police-focused AI systems because interpreting images is notoriously unreliable. There are countless ways to describe colors, facial expressions, or visual scenes, and different AI systems will produce wildly different results from the same prompts.
Where this is heading
Kinsey described the current situation as departments “building the plane as it’s being flown” with AI. Law enforcement typically waits until technology is already being used—and mistakes are being made—before creating guidelines. Some states like Utah and California are starting to require labeling when AI generates police reports, which seems like the absolute bare minimum.
So what’s the solution? Experts suggest starting with transparency and understanding risks before deployment. But with AI tools becoming increasingly accessible, the cat’s already out of the bag. The real question is whether law enforcement agencies will get ahead of this technology or keep playing catch-up while potentially compromising cases and people’s rights.
