Spain's opposition PP deepfake ad shows the real AI election threat: parties normalising fakes, not foreign trolls

Spain's conservative People's Party (PP) has published a campaign video that uses deepfakes, meaning AI-generated imitations of real people's faces and voices. The targets are figures linked to Prime Minister Pedro Sánchez's Socialist party. According to The Guardian, the video appears to show them admitting to acts such as soliciting prostitutes and taking kickbacks. It reached nearly 2 million views by Friday, a few weeks before next month's general election.

The Guardian describes two examples. An AI version of Sánchez's wife, Begoña Gómez, talks about getting a university post thanks to her husband's job. A deepfake of former Socialist prime minister José Luis Rodríguez Zapatero talks about hiding jewellery in a safe. Both lines refer to ongoing investigations, and both Gómez and Zapatero have denied any wrongdoing. The video carries a notice that it was made with AI, but in tiny, nearly transparent font.

The response came quickly. Óscar López, Spain's digital transformation minister, called the video a "colossal lie" and said it was "intolerable" for a party to post it on its official accounts. He also warned that anyone not used to AI could easily mistake it for reality. Leire Díez, a former Socialist party member whose likeness was used, filed a court complaint asking for the ad to be taken down. According to reports, she accuses the PP of defamation and digital identity theft. Even so, most of the nine legal experts consulted by El País believe the video is not illegal, because it carries an AI label and reads as political satire. Criminal law professor Juan Antonio Lascuraín drew the distinction that matters here: legality is one question, and whether the ad is ethical or exemplary is another.

Our thesis is that the most serious AI threat to elections is not the anonymous foreign troll farm. It is mainstream parties normalising the technique. Political scientist José Javier Olivas calls the ad a "red line" and "disinformation", because it puts words in real people's mouths. The satire argument assumes that "everyone knows" those people never said those things. But a deepfake is persuasive precisely because it looks real. An almost invisible label disappears when the clip is cut, reshared or watched on a phone. The likely path is easy to see: what starts as satire on an official account ends up as a clip with no context on WhatsApp.

This also fits a pattern we already track. We have argued that AI's real short-term risk is the automation of deception, not a distant superintelligence. The Guardian notes that AI-generated political ads are also a talking point in the US ahead of next month's midterms. When a large, established party in a European democracy uses the tool, it gives everyone else permission to do the same.

Our reading: this is a transition problem, and it can be solved. It does not call for panic, and it does not justify banning AI in political communication. We need rules that work in practice. AI labels should be visible and should travel with the content, not sit in fine print. Using a real person's likeness should require consent, or at least come with extra safeguards in campaigns. Platforms and electoral authorities also need to be able to act within days, not after election day. Synthetic-media detection and provenance tools (which record where a piece of content came from) are improving fast, and in the long run the same technology can help verify what is real. The risk, as with other AI regulation, is that lawmakers respond to scandal instead of designing evidence-based rules. Spain now has a clear test case. If the courts and parties set a firm limit, voters benefit. If they don't, the next campaign will treat this as normal.