AI deepfakes in the US midterms: the fake videos voters saw, and why most carried no label

In the run-up to the US midterm elections on 3 November, voters came across some unusual political videos. Politicians appeared to make statements they never recorded, and others were shown in conversations that never took place. The videos were made with artificial intelligence, and some looked surprisingly real.

A speech that was never filmed

One example came from the Texas Senate race. Republicans used AI to show Democratic candidate James Talarico speaking to the camera. His words came from his old social media posts, but he never filmed the speech. The ad carried a small AI label that was easy to miss, Reuters reported.

A conversation that never happened

In New York, Republican candidate Bruce Blakeman's campaign released a video showing Governor Kathy Hochul and Mayor Zohran Mamdani walking and talking together. Neither had taken part in that conversation. Democrats filed a complaint because the ad did not say AI had been used, the Associated Press reported.

$137 million, mostly unlabelled

There is serious money behind these ads. The Wesleyan Media Project, which tracks political advertising, counted 220 ads involving AI, with more than $137 million spent on them. Only 27% carried a disclaimer saying AI was used.

That means most viewers who saw one of these ads were never told that what they were watching was not real.

Did it change any votes?

Nobody knows yet. There is not enough research to say whether these videos changed anyone's mind about how to vote.

What is clear is that AI has become part of ordinary campaigning in the US, and that many of these videos left voters to work out for themselves what was real and what was not.