Former President Donald Trump posted what appeared to be a series of AI-generated images over the weekend to drum up support for his presidential candidacy, including a false endorsement from pop star Taylor Swift.
The posts show how Trump might wield generative AI in a way that confounds attempts to police AI-created election disinformation, thanks in part to long-standing legal precedent allowing candidates to lie in political ads. They arrive on the heels of Trump falsely accusing his opponent, Vice President Kamala Harris, of using AI to generate a rally crowd.
The images Trump posted included one that appears to resemble Harris from the back as she addresses a crowd in Chicago, the site of the Democratic National Convention this week, with a communist hammer and sickle dominating the background.
Another post includes screenshots of other users’ posts depicting images of “Swifties for Trump,” along with what seems most clearly to be an AI-generated image of Taylor Swift dressed as Uncle Sam with the words, “Taylor wants you to vote for Donald Trump.” Alongside the compilation of screenshots (only one of which was labeled as satire), Trump wrote, “I accept!”
Trump’s posts likely wouldn’t be covered by a growing list of state laws against election deepfakes, says Robert Weissman, copresident of Public Citizen. While roughly 20 states have enacted regulations around the use of AI-generated false images in elections, they typically ban depictions of someone doing or saying something in a convincing way. “So not just that it’s a well-done deepfake or well-done generative AI application, but it actually has to be plausible,” Weissman says.
Meanwhile, in terms of election disinformation, “there are no federal restrictions on the use of deepfakes,” Weissman says, noting the exception of the Federal Communications Commission’s ban on robocalls that use AI-generated voices. The nonprofit consumer advocacy group has been trying to get the Federal Election Commission to limit candidates’ ability to misrepresent their opponents with AI, but its rules likely wouldn’t cover something as obviously exaggerated as the Harris image or something that doesn’t depict an opposing candidate, in the case of the Swift image.
Still, Weissman suggested, Swift might have a claim for the use of her likeness to falsely issue an endorsement, perhaps under California’s Right of Publicity, protecting the use of a person’s likeness. Universal Music Group, which represents Swift, did not immediately respond to a request for comment about the use of her likeness in Trump’s post. The Trump campaign also did not immediately respond.
Courts have repeatedly determined that the First Amendment should often protect even deliberate lies, including by political candidates. If Congress passed regulations around AI deepfakes, that wouldn’t stop many uses of them. Weissman says an aggrieved opponent would need to show a lie led to harm or injury to voters, for instance, to make it illegal. “I don’t think that our legislative solutions are going to be perfect, even if we get exactly what we wanted,” he says.
Private platforms could take action against misleading generative AI content, of course, without government intervention. X’s synthetic and manipulated media policy prohibits such posts “that may deceive or confuse people and lead to harm.” But the platform has seemed to selectively enforce the rule, with even its owner, Elon Musk, appearing to violate it by posting a deepfake of Harris not clearly labeled as parody. Trump’s other venue of choice, his own platform Truth Social, has minimal rules in its community guidelines.
“It’s convenient for Trump, who was going around calling everything fake before AI, and wants us to call true things fake — like Harris’ crowds — to spread AI garbage to undermine the very idea of authenticity, and even reality in some ways,” says Weissman. “It’s very hard to have a democratic society if people can’t believe the things that they see and hear with their own eyes.”