AI-generated political ads flood midterm race and test platform labeling

Political campaigns are running ads built from obvious AI-generated images and video. In the midterm cycle, Forward Future has called the flood slopaganda, and it is testing how fast Instagram, YouTube and X can label synthetic political content before voters take it for evidence. CNBC reported that more than $187 million has been spent on nearly 300 ads generated or enhanced by AI, citing an analysis released last week by the Wesleyan Media Project. The organization says that figure is almost certainly an undercount, because many ads carry no disclaimer and determining AI use is hard in some cases.

The spending is lopsided. According to the Wesleyan Media Project report as CNBC described it, more than 90% of that money came from Republican candidates, a Republican party committee or a Republican-leaning group, and most of the ads did not include disclaimers about AI generation. CNBC found AI-generated ads running online without disclaimers, including a fake video of New York Democratic Gov. Kathy Hochul posted by her opponent, Republican challenger Bruce Blakeman. In Texas, Republican Ken Paxton's campaign created an AI-generated video showing Democratic opponent James Talarico making controversial statements. A three-second disclosure flashes in the middle of it, and the campaign is sharing it as organic content on YouTube and X. Organic content, as CNBC defined it, means free, unpaid posts on a profile or channel. The Blakeman and Paxton campaigns did not immediately respond to requests for comment.

YouTube told CNBC it applies prominent labels to realistic AI content. It requires disclosure when an election ad contains synthetic or digitally altered people or events, and it lets political candidates request removal of organic content where their face appears altered or generated by AI. Google also requires political ads to complete its election ads verification and carry paid-for disclosures. X said advertisers must comply with applicable laws, which places the responsibility on the advertiser rather than the platform. Meta pointed to its political ad guidelines, including restrictions in the week before an election, its paid-for labels and an ad library. It said that in June it added an about-this-ad tool for ads it detects were made with third-party AI tools.

The legal backdrop has no comprehensive federal rule. CNBC reported that no comprehensive federal law governs the use or disclosure of AI in political ads. Some 30 state laws set different requirements, and candidates don't always follow them. Two bills are named in the coverage. The Protect Elections from Deceptive AI Act, backed by senators from both parties, would prohibit knowingly distributing materially deceptive AI-generated video or audio about federal candidates. The AI Ads Act would expand existing bans on campaign misrepresentation to cover generative AI content. CNBC reported that none of the proposed bills have advanced out of committee.

The term itself has an origin. CNBC reported that slopaganda dates to early last year, when a group of AI experts used it in a paper describing generative AI slop combined with propaganda and misinformation. The generation side is also not a fixed barrier. CNBC cited a study from August by the Brennan Center for Justice that tested ChatGPT, Gemini, Grok, Claude, Perplexity and DeepSeek. It found that many models may avoid familiar conspiracies but can generally be manipulated into producing misleading material. CNBC also cited an October retest in which ChatGPT had begun redirecting prompts for harmful election images into comic-style illustrations or refusing them outright. The coverage reports that October finding only for ChatGPT, so it does not establish how the other models now perform.