The mayor of New York and the state governor stroll together through a sunny park, then garden side by side. They extol a new drug: a "pill against political delusion".
The scene never happened. It is an AI-generated advert, published by an opposing candidate, with no indication that it is artificial.
What is happening in the United States
According to data from the Wesleyan Media Project and the firm AdImpact, reported by the New York Times, more than 137 million dollars were spent on broadcasting political adverts made or retouched by AI during the midterm election campaign. Around 70% of them do not disclose it.
The reason is simple: cost. According to an advertising professional quoted by the newspaper, a spot that used to cost nearly 2 million dollars can now be produced for around 200,000 dollars.
And no camp is safe. A Republican incumbent defeated in a primary recounted that AI-generated adverts, funded against him, claimed he was dating Democratic elected officials. Some US states require a disclosure for "deceptive" content, others do not. A complaint has been filed in New York.
And in France?
France elects its president in April 2027. The rules there are very different, and markedly stricter.
In France, the Electoral Code bans all commercial advertising for electoral propaganda purposes, in the press, on radio, on television or online, during the six months preceding the month of an election. American-style paid spots therefore have no place there during a campaign. And since October 2025, a European regulation requires that all political advertising be clearly labelled, with the name of whoever paid for it. Several major platforms have even chosen to stop running it in the European Union. But these rules do not cover videos shared for free, which can circulate with no oversight at all.
That is where the real risk lies for 2027: not paid adverts, but videos and images made by AI, shared from account to account, with no way of knowing where they come from. We saw this with these AIs that create thousands of fake accounts on their own.
Five reflexes to avoid being caught out
1. Ask yourself who is speaking. Where does the video come from? A candidate's official account, a known media outlet, or an anonymous account created three weeks ago? Legal political advertising must state who paid for it.
2. Be wary of anything too enormous. An opponent caught in an absurd or shameful situation, a scandalous statement that confirms exactly what you already thought: that is precisely what forgers are trying to produce.
3. Watch and listen carefully. A voice slightly too smooth, lips slightly out of sync, strange hands, a blurry background. But be careful: fake videos are getting better and better, and the absence of any visible flaw proves nothing.
4. Check elsewhere. If a candidate had really said or done something that striking, the mainstream media would be talking about it. Go and look. Our five reflexes for dealing with a dubious photo apply to videos too.
5. When in doubt, do not share. A fake video only has power if it circulates. Every share gives it strength.
What to take away
The American campaign shows what an election becomes when making a fake video costs almost nothing. France is better protected against paid advertising, but not against videos circulating freely.
For 2027, the best protection will remain the oldest one: knowing where a piece of information comes from before believing it. We have detailed how to prepare your vote with AI without being taken in.