Pamela Evette still remembers the first time she saw herself doing something she had never done. In a dim, claustrophobic room, she sat under harsh lighting with sensors attached to her body, her face drawn and anxious. A polygraph machine hummed beside her, tracing jagged lines on paper. Then, as she watched, the machine burst into flames. The video was a political advertisement, designed to make her look deceptive and desperate. In small print, it said the dramatization was generated with artificial intelligence. But Evette, South Carolina’s lieutenant governor, quickly learned that small print was not enough. She told Noticias Telemundo that at first she dismissed it, thinking, “This is ridiculous, right? Like, nobody could possibly believe this?” It felt absurd. But when she traveled the state as a candidate for the Republican gubernatorial nomination, voters kept approaching her with genuine concern. “Oh my gosh, did they really make you take a lie-detector test?” they asked. She would patiently explain that no, it never happened, that the entire scene was fake, that the ad was a deepfake created by unseen political operators. What began as a laugh gradually became something heavier. “So you almost think it’s a joke and you kind of laugh at it, until you realize that there is a group of people out there that can’t tell the difference,” she said, “and then start to question if what they’re saying may be true.” The moment crystallized a new reality: in modern politics, seeing is no longer believing. Evette, who lost the June primary to Attorney General Alan Wilson, had been confronted by a machine built to make fiction look frighteningly real.
Her experience is not a one-off or an early-adopter curiosity. It is part of a wave of AI-fueled political advertising sweeping the nation. The Wesleyan Media Project, a group of experts that tracks political propaganda in campaigns, has identified at least 220 advertisements that were produced or enhanced with artificial intelligence by candidates or political action committees this election cycle. The total cost of those ads was just under $137 million. To understand how abrupt the shift is, consider that during the 2024 presidential election cycle, AI ad spending was less than $50,000. By May of this year, spending had already exceeded $20 million. The summer brought an explosion. “We really saw the growth in AI ads over the summer,” said Travis Ridout, a professor at Washington State University and one of the project’s directors. “A lot of them were connected with closely fought primary races.” These are not simple online banners. They are lifelike clips that can place a politician in a room that never existed, put words in a mouth that never said them, or cast an opponent in a disorienting, damaging light. The cost savings are enormous, and so are the political stakes. For every candidate who wakes up to a fake image of themselves, there is another campaign calculating how far the medium can go. Ridout suggested that candidates with smaller budgets are especially drawn to AI because it lets them compete with rivals who have larger campaign war chests, producing polished, persuasive work without expensive studios, actors, or editing teams. The technology is no longer experimental; it has become a standard tool in the political ad-maker’s kit.
South Carolina’s primaries showed how quickly the barrier between truth and counterfeit can erode. According to Evette, two AI-generated videos of herself were circulated during the campaign, paid for by a political action committee. Then, in a twist that underscores how normalized the technology has become, Evette’s own campaign released a video that falsely depicted Wilson rejecting a phone call from President Donald Trump. That video came from the candidate’s campaign, but it came with a disclosure: “This ad is AI-generated.” A request for comment sent to Wilson was not immediately returned. The back-and-forth shows that AI is not only being used by shadowy outside groups; candidates themselves are weaving it into their communications strategies. In other states, the pattern is repeating. During the Minnesota primary, for example, a deepfake circulated showing Lt. Gov. Peggy Flanagan, the Democratic Senate nominee, smiling in front of mountains of cash. Her office said the video did not stop her from winning, but it ignited alarm. “There is already too much distrust in our politics, and voters deserve to know what’s real and what’s not,” a spokesperson said. In Minnesota—one of the states that has tried to regulate the technology—the ad also produced a legal and philosophical debate: was it unlawful manipulation, or protected speech under the First Amendment? “Artificial intelligence should never be used to manipulate voters,” Flanagan’s spokesperson said. The question is no longer whether this technology will appear in elections. It is already here, weaving itself into races across the country, often with no clear signal to the person watching. In response to a query from Noticias Telemundo on social media, nearly 40 people said they had seen political ads generated with artificial intelligence, and some admitted that they have difficulty distinguishing them from the real thing.
Experts see both creative promise and real danger in this new landscape. The low cost and ease of AI tools may help level some playing fields, allowing lesser-known candidates to create advertisements that would otherwise be out of reach. Evette herself said she believes AI is here to stay, in politics and beyond. Campaigns “are extremely expensive,” she noted, and AI reduces many recording and editing costs. Ridout also said the technology could “unlock” more creativity in political advertising, which can often be dry and repetitive. The possibilities are practically endless, he said, and they include satire. In the ads analyzed by the Wesleyan Media Project, he saw candidates emerging from a swamp, boxing against Donald Trump, or appearing alongside physicist Albert Einstein, who died in 1955. But there is a darker side to those clever images. “I do worry about the lasting impact of those images, even if people know that it might not be true,” Ridout said. False images can still lodge themselves in a voter’s memory, shaping perceptions when they enter the voting booth. They can also generate waves of misinformation, feeding confusion and anger that spill beyond the internet and, in the worst cases, trigger violence. The very humor of a satirical image can make a lie easier to swallow, because a joke is often remembered long after its punchline is forgotten. Ridout warned: “And that could be deceptive; voters could be basing their decisions on falsehoods in that case. That’s, I think, what’s most concerning from a democratic perspective.”
Despite the explosion in use, the law has not kept pace. There is no federal statute specifically governing AI-generated political ads. For now, the responsibility rests with local governments. Ridout estimates that about 30 states have some type of rule or requirement regarding artificial intelligence in political campaigns. Yet of the 220 ads analyzed by the Wesleyan Media Project, only 27 percent revealed that they used AI tools. That means the vast majority of voters were left to decide for themselves, with no signal about what they were watching. Mekela Panditharatne, legal adviser and director of the Elections and Government Program at the Brennan Center for Justice, said existing state laws have not had as strong a deterrent effect as reformers would like. She emphasized the need for a universal federal standard. The Brennan Center, a nonprofit institute whose experts have analyzed the technology’s impact on democracy, wrote that AI can “redefine” modern political campaigns. Panditharatne said that without disclosure, generative AI in political ads “has the potential to distort perceptions of candidates, painting candidates in a flattering light or their opponents in a negative light.” In Congress, at least two federal bills are pending discussion. One would require political advertisements to contain conspicuous notices when artificial intelligence tools have been used. The other would attempt to prohibit using AI to generate materially deceptive content that falsely depicts federal candidates in political ads meant to influence federal elections. Lawmakers must weigh any restriction against the First Amendment, which protects freedom of speech, religion, and the press. But for many, the tradeoff is becoming clear: a little less illusion may be worth preserving the democratic faith that voters can know what is real.
In the end, it was not the election loss that haunted Pamela Evette. It was the moment when a fake version of her life started to feel real to the very people she was trying to reach. She no longer sees the lie-detector video as a harmless joke. It is a warning about what happens when technology outpaces the human ability to detect deceit. Evette said she believes AI is here to stay, and that campaigns will keep using it because it is cheap, fast, and endlessly adaptable. But the cost of accepting a world where any image can be faked is harder to calculate. When a political ad makes a candidate look like a liar, and enough people believe it, an election can become a contest of illusions rather than ideas. Voters need more than new laws; they need a shared standard of honesty, and they need technology companies, candidates, and the press to help them tell the difference. As Evette put it, there are people out there who cannot tell the difference between what is real and what is generated. That is not a joke. For a democracy built on informed consent, it is one of the most serious dangers of the digital age. The fire in the polygraph room may have been a computer-generated special effect, but the alarm it set off is entirely real.