One advertisement draws inspiration from Austin Powers, portraying U.S. Rep. Becca Balint (D-Vt.) as Dr. Evil and Molly Gray, a Democrat running for lieutenant governor, as Mini-Me.
Another video, based on Indiana Jones and the Temple of Doom, depicts an independent challenger of Balint ducking arrows and boulders while running through the “Temple of Two Parties.”
A third includes a parody of N.W.A’s song “Straight Outta Compton” that features rap verses recorded by a trio of Republicans running for the state Senate in Washington County.
The spots are proof that AI-assisted political advertising has arrived in the Green Mountain State. So far at least, the Vermont version of this national trend looks somewhat different — more obviously fake and much, much sillier — than many experts had feared.
Earlier this year, Vermont’s lawmakers moved overwhelmingly to curb “deepfake” political videos — that is, those intended to trick viewers into believing that they are seeing actual events. The legislators’ concern was that deceptive content could unfairly sway an election. Instead, what has emerged in the run-up to the November midterms is a campy mix of cartoonish attacks and pop-culture rip-offs that appear intended to provoke more than to deceive.
In Vermont, those using AI imagery in the political arena tend to be long-shot candidates and conservative commentators who say the tech offers a cheap, DIY way to churn out compelling content.
In Vermont, those using AI in the political arena tend to be long-shot candidates and conservative commentators who say the tech offers a cheap, DIY way to churn out compelling content.
“AI, like many technologies, is a tool and, like any tool, can be used to harm people or help them,” said Suzanne Seymour, a 64-year-old massage therapist and stand-up comedian who used AI to create the Indiana Jones-themed video — one of two she has made in support of her independent candidacy for Vermont’s lone U.S. House seat. In her second AI-assisted spot, Seymour saunters into a saloon to challenge her opponents to a “duel of words.” Balint is shown as a blazer-clad donkey with a bob cut and the Republican nominee, retired U.S. Army officer Gerald Malloy, as an elephant in military uniform.
Filming such a video on a real set would have cost Seymour far more than the couple hundred bucks she said she paid a content creator for each of the two AI-assisted videos.
Still, while the tools are giving people a colorful new way to make political statements, they also can be employed to produce the type of elaborate personal attacks that experts worry could deepen the nation’s political polarization.
Curtis Hier, a retired history teacher who runs a Facebook page called Flip Vermont Red, has posted more than three dozen AI-generated memes targeting Democratic candidates. The crudely made posts tend to depict the incumbents in an unflattering light, rolling around in a pile of money or literally kicking a can down a road.
Hier, who previously worked on Republican campaigns in New York State, where biting attack ads are routine, said AI is helping him inject some sting into Vermont’s electoral scene.
“Our politics are boring,” Hier said. “We want to shake it up a little bit.”
Hier sees himself as part of a long American tradition of political provocateurs. He points to Thomas Nast, the 19th-century cartoonist whose pointed anti-corruption drawings helped topple the political machine of New York City’s Boss Tweed.
“I can’t draw, and I can’t produce videos — I’m not trained in that,” Hier said. “But I do have a vision for messages I want to get out, so I use AI.”
Manipulated imagery has long been a part of American politics, from cartoons to photos or footage altered in ways that make candidates look better — or worse.
“It’s not that we’ve never had Photoshop,” said Scott Babwah Brennen, director of New York University’s Center on Technology Policy.
What has changed, Brennen and other experts say, is how easy it has become to produce these images and how pervasive they are becoming as a result.
“We’re seeing a lot more as these tools become more accessible,” said Michael Franz, a professor at Bowdoin College who codirects the Wesleyan Media Project, which tracks political ad spending across the country.
The project has identified more than 220 AI-generated political advertisements nationwide so far during the midterm election cycle. As in Vermont, the bulk of the ads nationally have come from conservative sources: More than 80 percent were produced by Republican campaigns or Republican-aligned political action committees, and more than $137 million has been spent promoting them, according to the New York Times.
Franz said his team likely only caught a portion of the ads circulating across the country, as some can be hard to detect. Although 30 states now have laws requiring the disclosure of the use of AI in campaign ads, most of those measures, including Vermont’s, apply only to deepfake videos and include exemptions for satire or parody.
The limits of regulating AI ads became clear this summer when an AI video targeting Balint began circulating online. It was posted by Hank Poitras, a combat vet who has gained notoriety for filming arrests in Brattleboro and posting them online under the moniker “Planet Hank.”
The video portrayed an AI-generated version of Balint issuing arrogant-sounding, made-up statements in front of the Vermont Statehouse. They included, “Straws are banned, and if those farmers don’t like it, they can leave.”
The video concludes with Mark Coester, a logger from Westminster who at the time was seeking the GOP nomination for Balint’s seat, using a backhoe to drop her into a dumpster while she screams.
Balint responded to the video with a real one of her own, in which she accused its creator of trying to trick voters. “Fabricating my voice and image to put words in my mouth is a new low,” she said.
Soon after, Attorney General Charity Clark notified Poitras that he was under investigation for potentially violating the state’s new AI law. Her office offered to drop the probe if Poitras uploaded the video again, but this time bearing a disclosure that it relied on AI.
Instead, he sued on free-speech grounds, and a federal judge ultimately blocked the probe, ruling that the video would likely qualify for the satire exemption.
Poitras, who did not respond to a request for comment, has since posted several other AI-generated videos depicting Balint, including the one casting her as Dr. Evil. In that video, Poitras, dressed as Austin Powers, bursts into Dr. Evil’s lair and stuffs Molly Gray — Mini-Me — into a sack that he then smashes against the ground.
Hier, the retired high school teacher, followed the Planet Hank case closely and said he agreed with the judge’s ruling. But he said he also supports the intention behind Vermont’s AI law — to protect voters from deception — and takes care to design his AI-generated posts so that they are obviously fake.
“Deepfakes are morally wrong, and in Vermont they’re legally wrong,” he said. “I’m careful to comply with it.”
Although Hier’s posts may be legal, that doesn’t mean they’re harmless, argues state Sen. Ruth Hardy (D-Addison), who was the lead sponsor of the AI law.
Hardy herself is one of Hier’s most frequent targets, featured in more than a dozen of his AI-assisted posts. He has depicted her in a variety of unbecoming scenes: flying over Addison County, witchlike, on a broom; setting money on fire; and weeping on the Senate floor.
Hardy has avoided looking at most of the posts but hears about them from others. She said she has noticed a pattern in the AI content circulating in Vermont: conservative men targeting Democratic women, often with a misogynist bent.
Five of the six incumbent lawmakers Hier has targeted are women. They include Sen. Becca White (D-Windsor), whom Hier has denigrated for using profanity in a TikTok video and for posting selfies online.
Hardy said AI makes it easy for people to create imagery that they might otherwise hesitate to create on their own. She worries that even visibly false depictions could rile up someone enough to push them beyond online criticism and to real-life harm.
Hier denies singling out women for attack, noting that he’s supporting a female Republican candidate who is running in Windsor County, White’s district, where there’s an open seat. He has also gone after Dave Wolk, a former Castleton State College president who is running as an independent to represent Rutland County in the state Senate.
Brennen, of NYU, said many of the AI advertisements airing across the U.S. this election cycle contain “quite nasty attacks,” and that could deepen the hostility that many voters already feel toward the opposing party.
Although such animosity long predated AI, “that doesn’t mean it isn’t contributing to the problem,” he said.
Whether the AI vogue is politically effective is another question. Negative ads have always been unpopular in Vermont. Pairing them with fake imagery — at a moment when concerns over AI are growing — seems unlikely to change that.
Even Hier concedes that his posts have generated considerable blowback from those who see his content as little more than “AI slopaganda.”
His favorite target — Hardy — said the negative AI ads mainly seem to have rallied supporters to her side.
“What I have heard from people is, ‘This is awful, this is unfair, this is gross, and I’ve got your back,’” she said. ➆
The original print version of this article was headlined “Artificial Influence | AI-created imagery is proliferating in Vermont this election season. It’s readily apparent, though, that the images are fake”
This article appears in The Tech Issue • 2026.
Got something to say?
Send a letter to the editor and we'll publish your feedback in print!
