This mid-term election cycle has been dominated by “slopaganda”, which refers to campaign advertisements containing generated segments and images from AI. This has presented another challenge for Instagram, YouTube, and X when monitoring platform content.
The term “slopaganda” can be traced back to the beginning of last year, when a group of experts AI used it in a paper to describe a new phenomenon, namely the combination of generative AI “garbage content” with propaganda and misinformation.
Currently, there is no comprehensive federal law that regulates the use or disclosure of AI in political advertisements. Although there are approximately 30 laws in various states in the United States that impose different requirements for the use and disclosure of AI, candidates do not always comply with these regulations. On major internet platforms operated by Meta, Google, and SpaceX under Elon Musk, users are exposed to advertisements generated by AI, as well as posts shared by politicians that are also generated by AI, interspersed with real videos and clips.
According to an analysis released last week by the Wesleyan Media Project, over $187 million has been spent on nearly 300 advertisements generated or enhanced by AI. The organization stated that this figure is almost certainly on the low side, as many of these advertisements did not include disclaimers. The report found that more than 90% of the expenditures came from Republican candidates, Republican committees, or pro-Republican groups, and that most of these advertisements did not disclose any information.
CNBC has discovered that there are multiple advertisements on the internet that have not disclosed that they were generated by AI, including a fake video of Kathy Hochul, the Democratic Governor of New York State. This video was released by her opponent, Republican challenger Bruce Blackman.
In Texas, Republican Senate candidate Ken Paxton produced a AI generated video that showed controversial remarks made by his Democratic opponent James Taralico. A three-second disclosure notice flashed across the screen during the video. This video was shared by Paxton's campaign team on YouTube and X as part of natural content.
Blackman and Paxton's campaign teams did not immediately respond to requests for comment.
YouTube told CNBC that the company will add “prominent tags” to the realistic AI content.
"For the natural content on YouTube, political candidates can request the removal of any parts where their faces appear to have been tampered with or manipulated using AI," the company stated. "Regarding advertisements, we require disclosure when election ads contain images or events that involve synthesized or digitally altered individuals or events, in order to provide transparency to the audience."
YouTube Google, the parent company, also requires political ads to complete their election ad verification process and to attach the " paid for by " disclosure information.
X indicates that advertisers must comply with applicable laws, and the responsibility lies with the advertisers, not the platform. The company did not respond to the comment request from CNBC.
Meta mentioned its political advertising guidelines, which include restrictions on the use of such advertisements in the week before elections. The company also requires the inclusion of the "paid for by" tags, and maintains a political advertising database. In June of this year, Meta strengthened its labeling measures by adding a "About This Ad" tool for advertisements detected as being produced using third-party AI tools.
Last month, a court in New Mexico ordered Meta to pay a fine of $25 billion in a data privacy case, which stemmed from a scandal involving Cambridge Analytica that occurred nearly a decade ago. Cambridge Analytica is a political consulting firm that is alleged to have obtained information about 87 million Facebook users in an improper manner.
One risk that generative models like AI pose to tech platforms is that they may be required to create misleading images and videos. The Brennan Center for Justice tested ChatGPT, Gemini, Grok, Claude, Perplexity, and DeepSeek to see if they would be willing to assist in producing deceptive content. A study released by the center in August found that while many models tend to avoid generating content surrounding well-known conspiracy theories, they can still be manipulated to create misleading materials overall.
Retests conducted on these platforms in October revealed that there has been progress with ChatGPT. It has begun to redirect prompts regarding harmful election images to illustrations, making them look like comics, or simply refuses to take any action regarding these prompts.
Some national political figures are attempting to resolve this issue.
The "Preventing Deceptive AI Act" ( Protect Elections from Deceptive AI Act ), supported by senators from both parties, will prohibit the deliberate dissemination of substantially deceptive AI generated videos or audio about federal candidates. Democrats in Congress are also pushing for other bills, including the " AI Advertising Act" ( AI Ads Act ), which aims to extend existing prohibitions on false statements in campaigns to generative AI content.
None of these proposed bills have yet been advanced in the committee.
