Donald Trump on Sunday shared several AI-generated images of Taylor Swift and her fans expressing support for his presidential campaign, reposting them with the caption “I accept!” on his Truth Social platform. The deepfakes are part of a series of images generated by artificial intelligence that the former president has been spreading in recent days, toeing the line between parody and outright election disinformation.

The AI-generated images Trump shared over the weekend show a series of young, smiling women in “Swifties for Trump” T-shirts, and an image of Swift dressed as Uncle Sam encouraging people to vote for the Republican presidential candidate. Each image is a screenshot from X, formerly known as Twitter, and was originally posted by right-wing accounts with a history of sharing misinformation. Swift has not endorsed Trump.

Trump’s posts come days after he also shared an AI-generated image of Kamala Harris addressing a communist military rally at the Democratic National Convention, as well as a deepfake video of himself dancing with X owner Elon Musk, an endorser. Trump’s embrace of AI-generated imagery threatens to further obscure an already murky information ecosystem surrounding the 2024 presidential election in which the former president routinely promotes falsehoods and conspiracy theories.

Concerns about AI-generated content influencing elections have persisted during the recent boom in generative artificial intelligence, with researchers warning for years that the technology has the potential to make it easier to create disinformation campaigns and flood online platforms with low-quality content. AI-generated disinformation has been circulating around elections around the world, as videos and images are used to troll opponents, falsify endorsements, and create deepfake audio intended to damage candidates.

While Trump has been sharing AI-generated images over the past week, he has also falsely claimed that a real image from one of Harris’ campaign rallies was the product of artificial intelligence and that the well-documented event never happened. His claim reflected a concept disinformation researchers call the “lie dividend,” in which a rise in manipulated content leads to general skepticism of all media and makes it easier for people like politicians to dismiss authentic images, audio or video as fake.

While most AI image generators from industry mainstays like OpenAI or Microsoft have placed restrictions on what they can create, banning images of public figures and rejecting prompts for political images, some users have found workarounds for some AI models or switched to others that lack such safeguards. Musk’s Grok image generator, which debuted last week, is capable of creating a series of images based on prompts that similar tools would reject, and has led to a recent spike in AI content surrounding the election. This includes images of political leaders, celebrities, and copyrighted works, as well as sexualized and violent content.

Almost immediately after Musk released Grok’s AI image generator, deepfake images of Trump and Harris began spreading on X. Many news outlets also reported that the tool could generate images of Swift, notable given that AI companies faced intense backlash earlier this year after sexualized deepfakes of the pop star circulated widely on social media. Swift has not endorsed a presidential candidate, but fiercely criticized Trump in 2020 for “fanning the flames of white supremacy” and vowed to vote him out of office.

Other Republican groups have also experimented with sharing AI-generated imagery this election season, including Ron DeSantis’ campaign during his failed bid for the GOP nomination. The Florida governor’s campaign shared a fake photo of Trump hugging Anthony Fauci, who has been a frequent target of conservative attacks. The Republican National Committee also stirred controversy last year when it released a partially AI-generated attack ad against Joe Biden, depicting a hellscape following his hypothetical election victory.