BLUF: On October 19, Axios reported that generative AI is now producing realistic midterm attack ads, some without any disclosure, and that election law has not kept pace. It cites the Wesleyan Media Project, which counted 220 AI-generated ads and $137 million in spending through October 11, mostly from Republicans. The lesson reaches past campaigns. Any named person or brand can now be shown saying or doing something that never happened, in a format built to look like paid media. This is one outlet's reporting, which we have not independently audited. We hold high confidence that the practice is spreading and low confidence in its effect on voters.
What Is Reported
The figures below come from Axios and the sources it cites. We have not reviewed the underlying data.
- Volume. The Wesleyan Media Project found $137 million spent on 220 AI-generated ads through October 11.
- Law. Four states prohibit certain uses of AI-generated deepfakes in political ads, 27 more require some form of disclosure, and no federal law specifically regulates them.
- Georgia Senate. A super PAC aligned with Senate Republicans has run AI-generated ads showing Sen. Jon Ossoff getting beauty treatments. Axios says they do not disclose AI use, which Georgia law does not require. Ossoff called for them to come down.
- Michigan Senate. A super PAC used AI to depict candidate Abdul El-Sayed in a fabricated family photo and a fabricated board-game scene. Axios says the ad carries a written and spoken disclosure, and the group defends it as grounded in his public record.
Why Disclosure Is Not the Same as Protection
Two of the cases above differ on one point: whether the ad says it is synthetic. Even where a label exists, the image is what people remember. A disclosure helps a viewer who is paying attention. It does little for the clip that is cut, reposted and stripped of its label.
This sits next to the issue covered in Colorado's first deepfake disclosure case: the legal route is slow, and the narrative moves in days. For a person or company on the receiving end, the practical timeline is set by the platforms and the news cycle, not by a court.
What Is Not Established
- Effect on votes. Spending and ad counts do not show persuasion. See AI influence operations: more capacity, unproven persuasion.
- Completeness. The count covers ads the researchers and reporters found. Undisclosed ads are, by definition, harder to count.
- Legal status. Whether any specific ad violates a state law is a question for the courts. We make no finding.
What Any Named Person or Brand Should Do Now
- Keep a dated record of what you actually said. Video, transcript and calendar for major public statements. Proof that something did not happen is hard. Proof of what did happen is easy if you saved it.
- Name one decision-maker for a response. A fabricated ad loses hours to the question of who may reply.
- Decide in advance what triggers a response. Reach, press pickup, or a named customer or regulator asking. Not every fake deserves a reply, and replying to a small one can amplify it.
- Know the takedown routes. Platform impersonation and synthetic media reporting channels, the broadcaster's or ad platform's complaint desk, and counsel for state-law notices.
- Prepare a one-paragraph, plain-language statement. What the real position is, where to verify it, and what the synthetic content gets wrong. Leave the date blank until needed.
- Monitor for the claim, not only the clip. The same false line is often repeated across formats. See narrative threat monitoring.
Questions Leaders Should Ask This Week
- If a realistic fake of our CEO or founder aired tomorrow, who decides the response in the first hour?
- Where is our dated record of what leaders have said publicly?
- Which platforms and ad networks would we contact, and do we have those contacts?
Organizations that want this built into an operating plan can read about our narrative and reputational threat intelligence work.
Confidence and Caveats
High confidence that AI-generated political advertising is established and growing, given the spending figures Axios cites. Moderate confidence in the specific cases, which rest on one outlet's account and the parties' statements. Low confidence in any measurable effect on voters. We will update this note if primary material becomes available.
Frequently Asked Questions
How many AI-generated political ads have aired? The Wesleyan Media Project, cited by Axios, counted 220 AI-generated ads and $137 million in spending through October 11.
Is there a federal law on AI political ads? Axios reports no federal law specifically regulates them. Four states prohibit certain uses and 27 more require some disclosure.
What should a candidate, executive or brand do about a fabricated ad? Hold a verified, dated record of the real statement, name one decision-maker, and know the takedown routes before an ad appears.
Kronus Intelligence Group builds and operates custom intelligence infrastructure for organizations that cannot afford to be wrong. If your environment requires it, we want to hear from you.
Start a Project →