Fake celebrity endorsements created with artificial intelligence a weapon in the disinformation war ahead of the 2024 elections

Washington (CNN)– Taylor Swift did not endorse former President Donald Trump last weekend. Ryan Reynolds was not photographed wearing a pro-Kamala Harris t-shirt. And the Communist Party USA never endorsed President Joe Biden’s now-defunct campaign.

But these false claims about the 2024 campaign, and dozens of other posts with similar false endorsements, have spread on social media in the run-up to the election, according to researchers at the News Literacy Project, a nonpartisan educational group that launched a new database on Thursday which collects more than 550 unique cases of election-related disinformation.

The latest and most visible example of these false claims emerged on Sunday, when Trump shared a post on his Truth Social platform that contained images created with the use of artificial intelligence suggesting a wave of support from Swift fans who called themselves “Swifties for Trump.” In response to the iconic pop star’s implied endorsement, Trump wrote: “I do!”

Swift, who had already criticized Trump for “stoking the fires of white supremacy and racism” during his presidency, supported Joe Biden in 2020, but has yet to endorse any presidential candidate in the 2024 election.

While one of the images in Trump’s collage, which shows Swift dressed as “Uncle Sam,” was obviously doctored, another, which showed a young woman at a rally, appeared authentic.

The other images that appeared to show large groups of smiling supporters celebrating Trump contained some of the hallmarks of AI-generated images, according to Lucas Hansen, co-founder of CivAI, a nonprofit dedicated to raising awareness about the growing capabilities and dangers of artificial intelligence (AI).

These images appear “highly retouched, have great camera quality” and “everyone is very good looking,” Hansen says. The images use “strong bokeh and a blurred background,” common traits of AI-generated images.

The News Literacy Project launched its panel on disinformation on Thursday to raise awareness about viral falsehoods that it says pose an “existential threat to democracy” and are best examined through mass analysis of hundreds of examples rather than individual fact-checks.

The database, which will be updated regularly, tracks several categories of political misinformation — conspiracy theories, lies about candidates' political views and fake endorsements — but the group does not measure how many times these viral posts are shared.

About 1 in 10 viral messages analyzed by the News Literacy Project contained fake endorsements, according to data provided exclusively to CNN. Those messages described supposed endorsements — or public snubs — from celebrities such as NFL quarterback Aaron Rodgers, actor Morgan Freeman, he musician bruce springsteen and political figures such as former first lady Michelle Obama.

The messages invoking these four figures accumulated at least 10 million views, according to the database.

At times, researchers found that separate posts were circulating on social media simultaneously claiming that the same celebrity had endorsed and denounced a candidate, underscoring the chaotic and misleading environment users are encountering.

“As a general rule, if a celebrity is wearing a T-shirt with an explicit political message, it’s very likely to be fake,” says Dan Evon, director of the News Literacy Project.

The proliferation of fake endorsements on social media comes as tech platforms dismantle guardrails and moderation policies designed to curtail the spread of dangerous misinformation.

ThisThe changes have been more acute in Xformerly known as Twitter, after tycoon Elon Musk bought the company and dismantle internal teams working to stop the spread of election misinformation, and restore the banned accounts of prominent conspiracy theorists and extremists.

Experts say the problem has been exacerbated by X's chatbot, Grok, which has alreadyhas provoked the ire of electoral officials for spreading false information about Harris' eligibility in the 2024 election. Last week, X began allowing users to use Grok to create AI-generated images from text messages, unleashing an avalanche of fake content about Trump and Harris.

“Going forward, Grok is likely to be one of the primary sources for these types of images, as it produces high-quality images, is easily accessible, and was intentionally made to have a low bounce rate,” Hansen said, adding that he was able to use Grok to create “Swifties for Trump” images that closely resemble the ones Trump shared.

X did not respond to requests for comment on the creation of misleading images related to political candidates.

Meta, which owns Facebook and Instagram, also cut some staff from its election integrity teams, as previously reported by CNN. The social media giant said last year that “no tech company does more or invests more to protect elections,” and announced that it would require political advertisers toto reveal whether they used AI to create or alter any images.

But even as some online platforms scramble to label AI-generated images and fact-checkers debunk the latest viral lies, the deluge of fabricated material can have an impact.

“If you repeatedly see these falsehoods exaggerating a candidate's popularity, they may stick with you, even if you know they are illegitimate when you pass them by in your feed”Peter Adams, vice president of research at the News Literacy Project, told CNN.

While the emergence of publicly available artificial intelligence tools has made it easier to create misleading claims, such as many of the fake endorsements identified by the News Literacy Project, other fake images circulating on social media use rudimentary Photoshop techniques.

The use of AI to craft political disinformation “has not been as prevalent and pernicious as initially feared,” Adams said, adding that traditional methods of image and video manipulation remain, for now, “much cheaper but just as effective.”