US police say they mistakenly posted AI-generated drug seizure photo on Facebook

The Star

a day ago

The officer did not ask police leadership before using ChatGPT, but no one in the department knew the chatbot could alter or change images as it did, Goldberg said. — Pixabay
The Westbrook Police Department said it mistakenly posted an AI-generated picture of items from a drug seizure to its Facebook page after an officer asked ChatGPT to add a police badge to the photo.
The image, which was posted Sunday evening, showed a collection of white powder in plastic bags, a scale, spoon and other paraphernalia spread out on a countertop. Police said the items, including methamphetamine and fentanyl, were seized from a home on Brackett Street last week where six people were arrested.
Some aspects of the image prompted criticism from hundreds of commenters, who pointed out malformed text and muddled lines, which could indicate an image may have been generated by artificial intelligence.
In a follow-up post Tuesday morning, the department denied that AI had been used to generate the photo, saying the packaging may depict words in "gibberish" because the drugs are illicit.
"The packaging is most likely foreign, and it is possible that whoever made the packaging used AI to make a clearly knock-off package," the post read.
Hours later, the department appeared to have deleted both posts. It posted a new update with the unaltered photo around 1pm Tuesday, clarifying that the earlier image had been altered by AI when an officer attempted to add the department's badge to the photo using an app.
Capt Steven Goldberg said in an interview at the department Tuesday that the app used was ChatGPT, a chatbot that can also create images with generative AI.
"Unbeknownst to anyone, when the app added the patch, it altered the packaging and some of the other attributes on the photograph," the department wrote in the Facebook post. "None of us caught it or realised it."
When confronted with the public's concerns, department leaders had assumed the packaging did contain "gibberish text," as do many drug-related materials, according to the post. The unaltered image shows the text on the packaging was in English and some items had been obscured or removed entirely.
The officer did not ask police leadership before using ChatGPT, but no one in the department knew the chatbot could alter or change images as it did, Goldberg said. When the police department was met with hundreds of Facebook comments, the officer showed Goldberg the original photograph and they noticed the inconsistencies.
Goldberg said the mistake has prompted procedural changes so that AI will not be used for social media posts.
"We've now learned our lesson," he said. "We didn't realise it would do this. Now that we know it can, we don't want to deal with that at all. There's too much at stake with that." – Portland Press Herald, Maine/Tribune News Service