Political campaigns in the United States have crossed a threshold in the 2025 midterm cycle: AI-generated deepfake ads are no longer edge cases but a standard campaign tool. The Wesleyan Media Project counts more than 164 AI-based ads this season, backed by over $80 million in spending. Republican candidates and their supporters account for 80 percent of that output. The ads are not trying to fool anyone into thinking they are real — they are designed to be obviously fake, grabbing attention through absurdity while embedding political narratives. Researchers at NYU's Center on Technology Policy call this phenomenon "slopaganda" — a portmanteau of low-quality AI "slop" and propaganda. The term captures the operational logic: these ads trade factual accuracy for emotional and ideological imprinting. Bruce Blakeman, a Republican running for governor of New York, used AI to fabricate scenes of Mayor Zohran Mamdani and Governor Kathy Hochul gardening together and conspiring to raise prices. Neither person appeared in the ads. Blakeman defends the practice as satire in a long American tradition, noting the ads carry brief AI-generated disclaimers. The regulatory landscape is a patchwork with a gaping federal hole. Thirty-one states have passed laws requiring AI-generated political ads to be labelled, but no federal statute exists. First Amendment protections for political speech, including parody, create a structural barrier. Ilana Beller of Public Citizen argues there is still a need for federal law, while Stanford's Starling Lab suggests the focus should be on judgment rather than bans — candidates who deploy AI recklessly reveal something about themselves to voters. The San Francisco congressional race to replace Nancy Pelosi became a live case study. Democratic state Senator Scott Wiener released an AI chatbot that parodied his opponent Connie Chan, mocking her record, her accent, and her naturalized citizenship. The backlash was immediate. Wiener withdrew the chatbot, conceding it "missed the mark," but the damage to his campaign brand was done. Chan's team seized the moment to frame the race as a proxy battle between working families and AI companies — a potent frame in a city that hosts OpenAI and Anthropic. The extraction pattern here is clear: campaigns spend on attention, voters absorb distorted signals, and no institution has the authority or speed to intervene. The brief disclaimer labels function as legal shields for campaigns, not as genuine voter protections. NYU's Brennen identifies the core dynamic — candidates "know there is political power to be gained by getting attention," and AI deepfakes are the cheapest attention machines ever built. The deeper concern is not that voters will believe the deepfakes are real but that the cumulative effect degrades the shared information environment. When every campaign ad could be AI-generated parody, voters must do the forensic work themselves. Public Citizen advocates making AI labels as clear as food labels, but no enforcement mechanism exists to ensure compliance. Some candidates have publicly pledged not to use AI deepfakes, and Beller notes voters seem to reward them for it — a fragile market signal in the absence of regulation. The midterm cycle is a dress rehearsal for 2028. If $80 million buys 164 AI ads now, the next presidential cycle will see orders of magnitude more. The question is whether the regulatory vacuum gets filled by federal law, by platform rules, or by nothing at all — leaving the information environment to degrade further under the weight of cheap, attention-optimized synthetic content.
