The midterm election cycle has been dominated by “slopaganda,” or campaign ads featuring AI-generated clips and images, posing new challenges for Instagram, YouTube and X as they seek to police their platforms.
The origins of the term slopaganda date back to early last year, when a group of artificial intelligence experts used the term in a paper to describe a new phenomenon that combines generative AI slop with propaganda and misinformation.
There is no comprehensive federal law regulating the use or disclosure of AI in political advertising. Additionally, there are approximately 30 state laws with varying requirements regarding the use and disclosure of AI, which candidates do not always comply with. On the leading internet sites run by Meta, google and Elon Musk’s space xusers will encounter AI-generated ads and AI-generated posts shared by politicians, along with authentic videos and clips.
More than $187 million has been spent on nearly 300 ads generated or enhanced by AI, according to an analysis released last week by the Wesleyan Media Project. The group says this is almost certainly an undercount because so many ads do not include disclaimers and in some cases it is difficult to determine the use of AI. More than 90% of the spending comes from Republican candidates, Republican committees or Republican-leaning groups, and most of it does not include disclosure, according to the report.
CNBC found numerous AI-generated ads posted online without public disclosure, including a fake video posted by Bruce Blakeman, a Republican challenger to New York Democratic Governor Kathy Hochul.
In Texas, Republican Senate candidate Ken Paxton created an AI-generated video of his Democratic opponent, James Talarico, making controversial remarks, with a three-second disclosure flashing in the middle. This is shared as organic content on YouTube and the X by Paxton campaign. “Organic Content” refers to free, free posts shared on your profile or channel.
Blakeman’s and Paxton’s campaigns did not respond to requests for comment.
On the left, Republican Texas Attorney General Ken Paxton and Democratic state Rep. James Talarico are candidates for the Texas Senate.
Michael Gonzalez | Sergio Flores | Getty Images
YouTube told CNBC that the company applies “conspicuous labels to realistic AI content.”
“For organic content on YouTube, political candidates can request its removal if faces appear altered or generated by AI,” the company said. “Requires advertisers to disclose when election ads include synthetic or digitally altered characters or events to provide transparency to viewers.”
Google, YouTube’s parent company, also requires political ads to complete an election ad verification process and include disclosure of “paid for” information.
Company X states that advertisers must comply with applicable law and that responsibility lies with the advertisers, not the platforms. The company did not respond to CNBC’s request for comment.
Mehta cited guidelines on political advertising, including restrictions on the use of political ads in the week before elections. The company also requires “label payments” and maintains an ad library of political ads. The company enhanced its labeling in June, adding a new “About this ad” tool that allows Meta to detect ads created with third-party AI tools.
Last month, Meta was ordered by a New Mexico court to pay a $25 billion fine as part of a data privacy lawsuit stemming from the company’s Cambridge Analytica scandal nearly a decade ago. Political consulting firm Cambridge Analytica allegedly improperly accessed the information of 87 million Facebook users.
One of the risks technology platforms face when it comes to generative AI is being asked to create misleading images and videos. The Brennan Center for Justice tested ChatGPT, Gemini, Grok, Claude, Perplexity, and DeepSeek for their willingness to promote deceptive content. An August study found that while many models may avoid producing content about well-known conspiracies, they can be manipulated to create content that is commonly misleading.
When we retested these platforms in October, we found that ChatGPT had improved. They began redirecting harmful election image prompts to illustrations to make them look like cartoons, or refusing to take action against them altogether.
Politicians in many countries are trying to address this issue.
The Protecting Elections from Deceptive AI Act, backed by senators from both parties, would prohibit the intentional distribution of AI-generated videos or audio about federal candidates that are materially deceptive. Other bills, including the AI Advertising Act, also led by Congressional Democrats, would expand current prohibitions on misrepresentation in election campaigns and generate AI content.
None of the proposed bills have passed out of committee.
Spotlight: The challenge for AI is the lack of computing in the market
