AI political ads expand as platforms struggle over disclosure and model misuse

More than $187 million fuels AI-assisted political advertising

Political campaigns have spent more than $187 million on nearly 300 political ads generated or enhanced by artificial intelligence, according to reporting published November 3, 2026. The material has been dubbed “slopaganda,” reflecting growing concern that synthetic imagery and video can enter election advertising faster than platforms can label or contain it.

For generative-AI developers, this shifts content controls from a general safety issue into an election-critical workflow. Image, video, and audio generators now need to consider how outputs may be used to mislead voters, not merely whether they depict recognizable people or harmful activity.

Disclosure enforcement becomes a platform bottleneck

Unlabeled AI ads are flooding the midterms as political campaigns direct tens of millions of dollars toward generative-AI advertising, leaving technology platforms struggling to enforce disclosure rules. Campaigns have used AI-assisted material in key senate and governor races, while some ads have drawn ridicule from critics who call them “AI slop.”

The operational problem is no longer limited to generating a persuasive image. Platforms also need repeatable ways to detect synthetic media, determine whether an advertisement used it, and apply disclosure requirements consistently. A generator that cannot preserve provenance signals or accommodate campaign review may create friction for agencies even if its visual output is effective.

Manipulated models can produce misleading election material

The Brennan Center for Justice tested ChatGPT, Gemini, and other popular AI systems and found that they could be manipulated to create misleading election content. The result shows that a model’s refusal behavior cannot be treated as a complete control when prompts are engineered to bypass safety safeguards.

An October retest produced a more encouraging result for ChatGPT, which refused the relevant requests. That change suggests model behavior may shift as vendors update safety systems, making repeated evaluation necessary. The evidence does not establish the same retest outcome for Gemini or provide a complete comparison among the tested models.

For teams building election-related generation products, the practical need is an evaluation process that records how a model behaves under manipulated prompts, rather than relying on a single successful test.

Voters struggle to distinguish synthetic campaign advertising

Most voters interviewed by CBS News Atlanta said they could not tell when a political advertisement had been created with AI. The finding matters because disclosure labels only work if they appear before voters encounter the material and understand what they are seeing.

Synthetic campaign media therefore requires two coordinated interventions: platforms need to identify and label AI-assisted advertising, while model providers need to reduce their ability to generate deceptive election content. The evidence indicates that voters may not recognize synthetic work unaided, making visible disclosure and reliable detection central parts of the response.

Regulation becomes part of the content-generation decision

Nearly half of participants in a recent Brennan Center for Justice-funded study favored more regulation of prediction markets. The result came alongside reporting that connected election content, voting advice, and “slopaganda” as connected policy concerns, although the evidence does not specify a particular rule or deadline for generative-AI advertising.

This creates uncertainty for teams that build generators or campaign tooling. A tool that supports synthetic assets, persuasive copy, or targeted political messaging may need configurable safeguards and disclosure features as regulatory expectations develop. For developer teams, evidence handling, campaign review, and audit trails should be treated as product requirements rather than optional additions.

What creators and reviewers should check next

The immediate workflow is to test whether generated images and videos carry usable provenance information, whether platforms reveal disclosure labels consistently, and whether safety evaluations still reproduce the October refusal behavior reported for ChatGPT. Campaigns and publishers should also retain evidence of how synthetic assets were created and reviewed.

For developers, the evidence points to a concrete need for synthetic-media detection, provenance inspection, political-content labeling, prompt-manipulation evaluation, and campaign-disclosure checks. Those capabilities would help teams test whether generated material can move from production into advertising without losing its origin signals or compliance review trail.