More than $137 million has been spent on political ads built at least partly with artificial intelligence ahead of the 2026 midterms, and most of them never tell the viewer. The figure comes from the Wesleyan Media Project, which identified 220 AI-generated ads across 34 states and says the real total is almost certainly higher. As Gizmodo reported, only about 27 percent carried a disclaimer telling viewers the content was synthetic.
The scale of the jump is the part that should give anyone pause. Wesleyan's co-director noted that the 2024 cycle saw less than $50,000 spent on AI-generated ads. Two years later it is a nine-figure line item. Whatever hesitation campaigns once felt about putting a machine-made scene in front of voters has largely evaporated.
The spending is lopsided. Republican candidates and groups accounted for 81 percent of the ads Wesleyan flagged and 88 percent of the money behind them. One example cited in the research is a Blakeman campaign video that places New York mayor Zohran Mamdani and Governor Kathy Hochul in a fabricated scene, with no label. The campaign has argued that satire does not need a disclosure, which is precisely the grey zone that makes the labelling question so hard to settle.
That grey zone is the whole problem. A clearly absurd parody and a subtly doctored clip can be built with the same tools, and a viewer scrolling past at speed cannot always tell which is which. When roughly three in four of these ads arrive unmarked, the burden of working out what is real lands entirely on the audience, usually in the few seconds before the next video starts.
None of this is the same as proving any single ad fooled anyone. The harder, slower damage is to the baseline. When synthetic political video becomes ordinary, genuine footage gets easier to wave away as fake, a dynamic we examined when weighing how much real harm AI fakery is doing. It is the same erosion that has worried AI firms warning courts that video evidence can no longer be taken at face value.
Disclosure rules exist in some states and not others, and enforcement before an election is slow by design. With the midterms close, the likeliest outcome is that most of these ads run their course unlabeled, and the argument about whether that should be allowed happens afterward, when the votes are already counted.
Sources
- i. gizmodo.com
- ii. www.minnpost.com
- iii. www.nbcnews.com
Commentarii · 0