This year, roughly half the planet goes to the polls. In June we have the European Parliament elections, and Romania has several rounds of elections coming up. I am not writing about who should win. I am writing about something that concerns all of us, whoever we vote for: for the first time, anyone can fabricate an audio or video clip of a politician saying things they never said, in a matter of minutes.
In January I launched AI Expo Europe in Bucharest precisely because I am convinced AI will change the way we work and live. That is exactly why I think we have a duty to talk openly about its ugly side, too. A powerful tool cuts both ways.
What has changed since the last elections
Election manipulation is nothing new. Rumours, photos taken out of context, fake accounts: we have known them for years, and I wrote about them when discussing trust and fake news. What is different now is the cost. Two years ago, a convincing fake video took time, money and skilled people. Today, a voice can be cloned from a few minutes of public recordings, and generated images have become so good that after I tested AI image generators myself, I wondered how much longer we can believe what we see.
On 15 February, OpenAI previewed Sora, a model that generates video from text. It is not publicly available, and the company says it is testing it carefully. But the direction is clear: video, the last format we instinctively trusted as proof that “this really happened”, is entering the zone where it no longer proves anything on its own.
Examples that have already happened
These are not hypothetical scenarios. A few cases from recent months:
- In September 2023, just days before Slovakia’s parliamentary election, a fake audio recording attributed to a party leader circulated online. The timing was deliberate: close to the vote, when it is hard to debunk in time.
- In January 2024, voters in New Hampshire received robocalls with an AI-generated voice imitating President Biden, urging them not to vote in the primary.
- Also in January, sexually explicit AI-generated images of Taylor Swift spread on X before the platform reacted. Not an election case, but it shows how fast a fake can travel before anyone steps in.
All three share the same ingredients: speed, emotional charge and a short window in which the fake does its damage before it gets corrected. It is the same logic we see with fake profiles on platforms, only the stakes are much higher.
What we can do as citizens
You do not need to be an expert to protect yourself. You just need to slow down a little. A few simple rules I follow myself:
- Check the source. Who posted the clip first? An account created last week with no history is not a source.
- Be suspicious of emotional last-minute clips. If something makes you angry or scared and appears a day or two before the vote, that is precisely when you should be most careful.
- Look for confirmation from reputable outlets. If a politician really said something outrageous, several credible newsrooms would report it. If you find nothing but the clip, that is a red flag.
- Don’t share in a panic. Fakes live on our sharing. Waiting five minutes costs nothing, but it can break the chain.
- Talk to your parents and grandparents. Many of them don’t know a voice can be cloned. A chat over coffee or on the phone is worth more than any official campaign. The same trick is already used in “Mum, I’ve had an accident, send money” scams.
What platforms and AI companies should do
The burden cannot fall on users alone. On 16 February, at the Munich Security Conference, major tech companies including Google, Meta, Microsoft, OpenAI, TikTok and X signed an accord to combat deceptive AI-generated election content. It is a good step. But a voluntary accord is only worth as much as its enforcement.
What I would like to see in practice: visible labels on content generated or altered with AI; technical watermarks, hard to remove, added by the models that generate images, audio and video; fast removal of obvious fakes targeting the voting process; and full transparency on political ads: who pays, how much and whom they target. I have written before about free speech on social platforms, and I still stand by it. But the freedom to speak your mind does not include the freedom to put words in someone else’s mouth.
What governments and the EU are doing
Europe already has some tools. The Digital Services Act requires very large online platforms to assess and mitigate systemic risks, including risks to electoral processes, and to be more transparent about advertising. In December 2023, a political agreement was reached on the AI Act, which I covered last year when writing about how Europe is regulating AI. The text, still awaiting formal adoption, includes transparency obligations for deepfakes: content that is artificially generated or manipulated must be disclosed as such.
The problem is timing. Laws kick in gradually, and the elections are now. So in 2024 we will rely more on platforms, the press and everyone’s common sense than on penalties.
Journalists and companies have a role too
Newsrooms can do two simple things: verify before picking up a “viral” clip, and explain to their audience how they verified it. Fact-checking that shows its method educates rather than just corrects. Companies, including small ones, can protect their own voice and image: a clear official channel where every important statement is published makes it much easier to debunk a fake in your name. And if the CEO “calls” an employee asking for an urgent transfer, internal procedure should require confirmation through a different channel. It sounds trivial until the day you need it.
Don’t let fear make you stop believing anything
There is an opposite danger, just as serious. Researchers call it the “liar’s dividend”: once everyone knows recordings can be faked, anyone caught saying something real can simply claim “it’s a deepfake”. If we end up believing nothing, the liars win anyway.
The answer is not cynicism, it is verification. Have a few sources you trust, know what an official statement looks like, and accept that sometimes the right answer is “I don’t know yet”. Democracy survived the printing press, radio and television. It will survive AI too, if each of us does our part.
How do you check information during election season? Has a clip ever reached your family that turned out to be fake?