Imagine scrolling through your social feed and stumbling on a video of New York’s governor, Kathy Hochul, and the mayor of New York City, Zohran Mamdani, calmly gardening side by side, then cheerfully agreeing to raise prices on everyday residents. In another clip, Hochul proudly talks about hiking tolls and subway fares until ordinary people can no longer afford to drive or ride the train. It feels like a bizarre political fantasy—and that’s because it is. Neither Hochul nor Mamdani ever filmed those scenes. The ads were fabricated with artificial intelligence and released by Bruce Blakeman, a Republican candidate for governor of New York, as part of a campaign that has leaned heavily into what researchers are now calling “slopaganda.” That awkward mashup of “slop” and “propaganda” describes low-quality, obviously fake AI-generated content that is not designed to deceive so much as to grab attention and provoke laughter, anger, or shares. Scott Babwah Brennen, director of New York University’s Center on Technology Policy and co-author of a recent report on the phenomenon, explains that while deepfake regulation has mostly focused on labels that warn voters about realistic AI manipulation, the most defining feature of this election season is actually slopaganda—ads so clearly fake that no one could mistake them for real footage, yet they still dominate the political conversation. According to the Wesleyan Media Project, which tracks election advertising in a nonpartisan way, candidates and their supporters have poured more than $80 million into more than 164 AI-based ads this cycle, and Republican candidates and their backers are responsible for about 80 percent of them. The numbers reveal how quickly AI has become a standard tool in modern campaigning, not as a secret weapon but as an open, almost theatrical performance for voter attention.
Blakeman’s ads are a perfect example of this new genre. They carry a brief disclaimer at the start saying they are AI-generated, and Blakeman insists they follow a long American tradition of satire and parody in politics. He says voters find them humorous, and he is not wrong—at least in the sense that many viewers do share them for entertainment. But the humor is not innocent. Brennen points out that slopaganda is often used to “communicate deeper truths rather than factual truths,” which means these absurd images can shape what people believe about a candidate’s character or ideology, even when they know the footage is fake. If a candidate is repeatedly shown as someone who wants to raise prices or make life harder, the emotional impression may linger far longer than any fact-check. This season has produced a striking gallery of such images. A Democratic Senate candidate from Ohio, Sherrod Brown, was portrayed as the last guest at a party, long after everyone else wanted him to leave. James Talarico, a Texas Democratic Senate candidate, was shown wearing a blue dress reminiscent of the character Maria from The Sound of Music, while Michigan Republican Senate candidate Mike Rogers was transformed into a caped superhero. Each ad is transparently artificial, often goofy, and yet every one of them makes a pointed political claim—about alleged overstaying, about being out of touch, about being a savior or a joke. The effect is subtle: rather than lying about a candidate’s record, these ads plant an impression, a gut feeling that may prove more powerful than any policy paper.
Perhaps the most dramatic case of AI experimentation this season involved Scott Wiener, a Democratic California state senator running in the high-profile race to succeed House Speaker Emeritus Nancy Pelosi, who represents San Francisco in Congress. Wiener released an AI chatbot designed to answer questions about his opponent, Connie Chan, a fellow Democrat who served on the city’s Board of Supervisors. The chatbot was not a harmless novelty. It mocked Chan’s record on housing, allegedly portraying her as someone who blocked real estate development in the city, and it also parodied her accent and her naturalized citizenship. Within hours, the backlash was intense. Critics called it bigoted and xenophobic, and Wiener quickly withdrew the chatbot, saying that although he thought it was innovative, “it missed the mark.” The episode illustrates how quickly AI can cross the line from edgy satire to harmful stereotyping, especially in a city that is both the epicenter of the AI industry and a place with a large immigrant population. It also shows the dilemma candidates now face: using AI can get attention, but it also exposes their own judgment. Adam Rose, chief operating officer of the Starling Lab, a research center on technology and human rights founded by Stanford and the University of Southern California, says candidates are always making “judgement calls” when they choose to use AI in their ads, and those decisions reveal something important about them. In Wiener’s case, Rose says, releasing the chatbot “showed bad judgement. It showed voters something about him.” That may be exactly why some voters take notice of AI gaffes—not because the technology itself is evil, but because it exposes how candidates think about people, rivals, and the values they claim to defend.
The legal landscape around AI-generated political ads is still patchy. There is no federal law specifically regulating ads made with AI deepfakes, although 31 states have passed their own laws requiring some form of labeling. Ilana Beller, director of State AI Governance and Policy at Public Citizen, a consumer rights advocacy group based in Washington, D.C., says state laws can only do so much without running into First Amendment protections for political speech. “There is still a need for a federal law,” she insists, because voters in different states are being given wildly different levels of disclosure. Some labeling requirements are weak, while others are stronger, and in states without any law, AI ads may appear with no warning at all. But even among those who favor regulation, there is agreement that not every AI touch needs to be disclosed. Adam Rose argues that minor corrections, like adjusting color or fixing audio, are not the kind of manipulation voters need to worry about, and that banning AI-generated ads outright would be wrong. Instead, he says, the important thing is that candidates take responsibility for the messages they put into the world. A label can tell voters that an ad is synthetic, but it cannot tell them whether the underlying message is fair, honest, or respectful. That burden falls on the candidates themselves. Beller, however, wants labels to become as clear and easily understood as nutrition labels on food, so that ordinary voters—including younger people and seniors—know exactly what they are watching. She says people need to be taught to look for these labels, and that public awareness is crucial as AI becomes more common in politics.
The battle over AI ads is also playing out in the streets of San Francisco, where the stakes are uniquely high. The city is home to OpenAI and Anthropic, two of the largest AI companies in the world, and its representative in Congress will be at the center of future legislation on the technology. Polls show that a majority of Americans are more concerned than excited about AI, so the race between Wiener and Chan is being watched as a possible preview of how the country will handle the industry’s growth. Ian Krager, a spokesperson for Chan, framed the election as a choice between working families and tech corporations. “This race is a harbinger for the country,” he said. “It is a moment where we can show the world America can lead the world on AI development without asking working people to bear the cost.” He added that Chan is clearly aligned with working families, while Wiener is aligned with the AI companies. As a state senator, Wiener actually drafted a law last year that would have required AI companies to disclose security incidents involving dangerous or harmful behavior by extremely large AI models. The final version that passed was less stringent, but the effort showed that he was engaged with AI policy in a serious way. The chatbot incident, however, gave his opponents an opening to paint him as someone who is too cozy with the industry and too willing to use AI in careless, even harmful ways. It is a reminder that in the AI capital of the world, every use of the technology by a politician is likely to be scrutinized not only for its political effectiveness but for its ethical implications.
Looking ahead to the final stretch of the November 3 midterm elections, and beyond, the role of AI in campaigns is only going to grow. In New York, the Democratic Party has filed a complaint with the state board of elections, arguing that Blakeman’s ads violate disclosure requirements and are “misleading to the voters.” Blakeman, for his part, continues to defend them, saying that people enjoy the satire and that the ads are clearly meant as parody. Brennen, the NYU researcher, suggests that such ads are really about capturing attention in a crowded media environment. “Candidates know there is political power to be gained by getting attention,” he says, and AI is an incredibly cheap and fast way to generate wildly shareable content. Yet there are also signs of pushback. Some candidates have publicly pledged not to use AI-generated ads at all this cycle, and Beller notes that voters seem to appreciate that restraint. “People seem to like them for it,” she says, suggesting that honesty and caution might themselves become political assets. The challenge is that the attention economy rewards outrage and novelty, and AI is the ultimate creator of both. As voters head to the polls, they will not only be choosing between candidates and parties; they will also be navigating a media landscape where the old rules of truth and fiction have been bent into strange new shapes. The hope among reformers is that a combination of clear labels, public vigilance, and candidate accountability can keep the worst impulses in check. The fear is that by the time voters learn to recognize slopaganda for what it is, it will have already shaped their opinions in ways they may not fully understand. In that uncertain space, the most human thing a candidate can do is show good judgment—and the most powerful thing a voter can do is stay curious, skeptical, and aware that what they see may not be real, even when it makes them laugh.