AI fakes of accused US press gala gunman flood social media

Within hours of authorities identifying the suspect as Cole Tomas Allen, 31, of California, AI-generated images depicting him beside numerous celebrities pinballed across Facebook in posts saying he was their "former driver," "assistant" or "production crew member."

An AFP investigation found more than 50 public figures falsely associated with Allen, from actors Tom Hanks and Sydney Sweeney to musicians Chris Brown and Taylor Swift.

Politicians including former US president Barack Obama and Canada's Pierre Poilievre were also falsely implicated, as well as Pope Leo XIV and NBC News anchor Savannah Guthrie.

Meta did not immediately respond to AFP's request for comment.

The fakes reflect an online ecosystem saturated with content known as "AI slop." Once largely focused on celebrities, generative content has quickly scaled to portray individuals like Allen, whose online presence was limited.

"Two years ago, you probably wouldn't have been able to make those images of him, because we could only really make compelling fakes of celebrities who had a large digital footprint from which the AI systems had been trained," said the University of California, Berkeley's Hany Farid, who is also chief science officer at GetReal Security. "Now, all I need is a single image of you."

Aaron Parnas, an independent journalist whose likeness appeared in AI-enabled posts claiming Allen worked for him, pleaded on Facebook for people to report the "completely fake" images.

"This is extremely dangerous," Parnas told his followers.

- 'Designed for virality' -

A separate rush of posts falsely claimed Allen had been on staff for over 40 different professional and collegiate sports teams, with AI-generated visuals dressing him in gear for teams across the NFL, NHL, NBA, WNBA and NASCAR.

Many of the renderings appear based on the picture from a tutoring company's post recognizing Allen as "teacher of the month" in December 2024.

The template-driven format resembles the output of content mills that mass-produce made-up clickbait stories, said digital literacy expert Mike Caulfield.

"This looks a lot like the same content farm behavior, just with AI," Caulfield told AFP.

Recent improvements in AI technologies have made visual fakes easier to create and more convincing, with once-telltale mishaps such as six-fingered hands increasingly less common.

"AI makes it trivially easy to take existing photos and change their clothes, environment, or to swap out someone else's face," said Jen Golbeck, a professor at the University of Maryland’s College of Information. "As soon as someone gets an idea, they can make it a visual reality."

"Five years ago, it would not have been unusual to see people manually photoshopping pictures like the ones we are seeing, but it would never have been at this volume."

Researchers expressed fears about the quantity wearing on social media users, who could tire of determining what is real.

AFP documented similar bursts of fakes after other major events, including the US capture of Venezuelan leader Nicolas Maduro in January and Charlie Kirk's assassination last year.

"These things are being designed for virality, and then of course the algorithms pick up on them," said Farid, from GetReal Security. "It's super profitable."

"Every time there's a world event, we are just flooded with this kind of nonsense. I don't think that's going away."

bmc/mgs/pnb