Trump arrest photos fabricated using artificial intelligence

Copyright © AFP 2017-2023. All rights reserved.

Dozens of pictures shared on Twitter purport to show Donald Trump arrested and jailed on charges related to a hush-money case in New York City. But the photos are fake; the person who originally posted the images told AFP he created them using artificial intelligence, and no charges against the former US president had been made public as of March 24.

"HOLY SHlT THEY GOT TRUMP," says a March 21, 2023 tweet sharing two images of Trump being apprehended by police officers.

"These photos of Donald Trump getting arrested are too much," says another tweet posted the same day.

Collage of screenshots taken from Twitter on March 22, 2023

Pictures of Trump being arrested, imprisoned and wearing an orange jumpsuit accumulated thousands of interactions after the former president claimed March 18 that he would "be arrested on Tuesday of next week" and called for his supporters to "protest" and "take our nation back."

Trump's post on Truth Social referenced an investigation from Manhattan District Attorney Alvin Bragg that centers on $130,000 paid weeks before the 2016 election to allegedly stop porn star Stormy Daniels from going public about an affair she says she had with the real estate mogul years earlier. Michael Cohen, Trump's former lawyer, says he made the payment and was later reimbursed.

The payment to Daniels, if not properly accounted for, could result in a misdemeanor charge for falsifying business records. That might be raised to a felony if the false accounting was intended to cover up a second crime, such as a campaign finance violation.

But no charges had been announced as of 2000 GMT on March 24, and Bragg has said the former president created a "false expectation" of his imminent arrest.

AI-generated photos

Several of the images shared online stem from a March 20 Twitter thread from Eliot Higgins, founder of the investigative news outlet Bellingcat.

"Making pictures of Trump getting arrested while waiting for Trump's arrest," Higgins said in the first tweet of his thread. But many of those resharing the images made no reference to the fact that these were digital creations, and some comments suggested the users believed they were genuine.

Higgins told AFP he created the images using MidJourney, an artificial intelligence program. He said he was later banned from using the software.

"I don't have access to their services anymore," Higgins told AFP on March 22. "It started off as just playing around to see what it was capable of, then I strung together a few images in a thread that made a little story and it took off from there."

Higgins created 124 photos using MidJourney, according to a Google Drive link he shared with AFP. Several of the images match the posts shared online.

"Many people create AI-generated images with no intent to manipulate," said Renee DiResta, research manager at Stanford Internet Observatory. "In this case featuring the Trump images, the creator disclosed that they were AI-generated in his own post, but that context was lost when they were reshared."

AFP reached out to MidJourney and the Trump campaign for comment, but responses were not forthcoming.

How to spot fabricated images

AI-generated photos of public figures such as Trump and French President Emmanuel Macron have become increasingly popular online in recent weeks.

On March 23, Trump shared what appeared to be a picture of himself praying. AFP did not find the photo elsewhere online through a reverse image search, and US business magazine Forbes pointed out visual discrepancies within the photo.

Del Walker, a senior character artist for the video game developer Naughty Dog, told AFP that distorted faces and limbs, gibberish text, "hypersmooth" skin, and objects with odd features are all signs of AI manipulation. But he cautioned that such red flags "can be improved in future AI updates or easily edited in image editing software."

"The only real kryptonite for AI is skepticism and scrutiny of all images," Walker said.

DiResta of the Stanford Internet Observatory agreed.

"We're at an interesting technological tipping point where realistic AI-generated images of anything -- including people -- are increasingly easy for just about anyone to create," she said.

"Explaining the potential for decontextualization, and conveying to the public that they should be appropriately skeptical of sensational images shared by accounts they don't know, is information we can begin including in online media literacy."

AFP has previously fact-checked misinformation about Trump's potential indictment here and here.