Opposition research operates on the guiding principle that someone’s beliefs, influences, and actions tell us something meaningful about their judgment, priorities, and fitness for public office.
Artificial intelligence offers campaigns a convenient, if deeply unethical, workaround: Have a digital puppet of the opponent say or do whatever makes the strongest attack.
Pennsylvania Republican gubernatorial candidate Stacy Garrity recently defended releasing an advertisement starring an AI-generated Gov. Josh Shapiro. The synthetic caricature gleefully performs his leading role in right-wing fan fiction, at one point peering through a telescope at Epstein Island.
As I wrote in my previous post about campaign messengers, the person delivering a message is part of its meaning.
The editorial responsibilities only grow when campaigns manufacture portrayals of opponents from marginalized communities. Even without explicit instruction, generative AI systems can reproduce racial and gender stereotypes; researchers have also documented latent antisemitic stereotypes in language-model outputs.
Consider the implications for a Jewish politician such as Shapiro. At one point in Garrity’s advertisement, his generated image looks straight into the camera, hundred-dollar bills fanned out before his face as he coos, “I’m not sharing a dime.”
Depicting a Jewish politician clutching money invokes a recognizable antisemitic trope, regardless of whether its creators intended that association. These portrayals circulate against a backdrop of real violence; Shapiro and his family survived an arson attack at the Governor’s Residence after celebrating Passover in April 2025.
In Texas, the Houston Chronicle reported that American Principles Project made an eight-figure buy to promote an AI-generated advertisement based on the famous “Daisy” ad from 1964. In this version, an AI-generated image of Democratic Senate candidate James Talarico is a masked doctor looming over the little girl on an operating table as the narrator recites common GOP fallacies about trans children.
When campaigns choose to circulate images portraying gender-affirming care as a threat, the consequences extend far beyond whether an advertisement moves a poll number.
In 2022, major medical organizations warned that coordinated disinformation campaigns were contributing to threats against hospitals, physicians, and families seeking care. A woman subsequently pleaded guilty to making a bomb threat that forced Boston Children’s Hospital into lockdown.
Campaigns are already facing legal threats over fabricated speech. Last month, in Wisconsin, Rebecca Cooke’s congressional campaign sent Republican incumbent Rep. Derrick Van Orden a cease-and-desist letter over AI-generated videos falsely attributing extreme positions that she never expressed to her.
Campaigns should not need the threat of costly litigation to establish a professional standard. As discussed in my previous post about manufacturing legal documents to enhance visual impact, staff make editorial decisions every day about what the evidence supports, what a visual implies, and what they risk if an attack cannot withstand scrutiny.
Artificial intelligence makes fabrication easier, which only makes editorial judgment more important. A campaign is still responsible for every word it puts in someone else’s mouth.
