The release of millions of documents by the U.S. Justice Department related to the crimes of Jeffrey Epstein, often referred to as the “Epstein files,” has raised questions about what’s been redacted.
Many photos and documents from the files are covered with black boxes, with some people using AI in an attempt to reveal what’s behind the redacted sections. However, as a result, false information and fake images are being produced and spread.
Giancarlo Fiorella is the director of research and training at Bellingcat, an investigative collective. He told CTV News Channel on Thursday that Grok, the AI tool for the social media platform X, is being used to unblur and un-redact images related to the Epstein files. However, of the 31 requests made between Jan. 30 and Feb. 5 from people attempting to use Grok to unblur images from the Epstein files, 27 led to fake images of a compromised photo, Fiorella said.
“Like all other AI imaging platforms. Grok has no way of knowing what’s actually hidden by the redaction by the black box,” he said. “It’s making a guess, based on other clues in the image, on what a person might look like.”
In some cases, Grok refused to unblur photos. Fiorella said that in four cases, Grok responded to users by saying it wasn’t able to process their request, and in one case, it ignored the request altogether. In another case, Grok responded to the user by saying it couldn’t do what was requested because of legal and ethical reasons related to the nature of the case.
In two instances, Grok declined to act out the request, because it told the users they weren’t subscribers to the platform’s premium service.
“Grok is an interesting platform, and it has all kinds of reasons for why it answers questions in that way,” said Fiorella.
Another consequence of AI tools is the creation of fake images, and the Epstein files were not spared. Many phony photographs have been doctored using AI to show Epstein with world leaders, like Mohammed bin Salman Al Saud, the Crown Prince of Saudi Arabia, and New York Mayor Zohran Mamdani as a child.
“Unfortunately, like in any other high-profile case with the advent of these sort of AI image generated platforms, we’re seeing a lot of fake images that look very realistic,” said Fiorella. “Some of them are indistinguishable from real life, and again, are complete fabrications.”
He explained that there are a wider range of tools easily available online that can create these realistic, but fake images.
“All you need to do is type a text description of what you would like to see,” he said.
As for why people do this, Fiorella said there’s a slew of reasons, some more sinister.
“We could be talking about even something like state actors who are attempting to create disinformation in order to sway public opinion in a country, or just generally confuse people,” he said.
Fiorella’s team contacted X regarding the fake images, but have yet to hear back.
