Opposition criticised for using AI-generated deepfake voice of PM in Polish election ad

Poland’s largest opposition party, the centrist Civic Platform (PO), has aroused controversy by broadcasting an election campaign advert that mixes genuine clips of words spoken by the prime minister with fake ones generated by AI.

Only a few hours following publication of the video did PO clarify that some of the material was artificially generated, after it had faced widespread criticism from commentators and experts for initially failing to do so.

“They call themselves the United Right,” began PO’s advert, referring to the name of the national-conservative ruling coalition. “But you probably know how it really is. And what do you think of the prime minister? Listen for yourselves.”

The video then alternated between showing genuine video clips of Prime Minister Mateusz speaking and audio clips of a voice that sounded like Morawiecki’s reading sections of leaked emails purportedly from the inbox of the prime minister’s former chief of staff Michał Dworczyk.

While the government has admitted that Dworczyk’s inbox was hacked, it has claimed that some of the material subsequently leaked online is fabricated. However, it has refused to confirm which individual emails are genuine and which are false.

PO’s aim in juxtaposing Morawiecki’s public statements with the content of the emails was to show that, while he was proclaiming the unity of the United Right in public, his private messages behind the scenes showed bitter tensions, especially between him and justice minister Zbigniew Ziobro.

However, nowhere in the video was it noted that the voice reading the quotes from the leaked emails was not Morawiecki himself, but an AI-generated voice designed to sound like him. That led many experts to criticise the party.

“When a voice clone is made, it should be marked in the post itself to make it clear that this recording was made using AI,” tweeted Aleksandra Przegalińska, a professor at Kozminski University in Warsaw specialising in new technologies.

“We appeal to all political parties and politicians to comply with…the principles set out in a recently adopted EU regulation, the AI Act,” wrote fact-checking website Demogog. “Any AI-generated content should be clearly marked. A note in the subsequent tweet is not enough!”.

Yesterday evening, a deputy interior minister, Błażej Poboża, told broadcaster Polsat that there are “risks associated with using these kinds of tools”.

“If we create false statements during the campaign, which are spoken by a voice very similar to a politician because it was created using artificial intelligence, it may be used for various provocations in the future,” he added.

A senior PO figure, Borys Budka, defended his party’s advert as a “very good idea”. Asked by Polsat why there was no information given that an artificially generated voice was used, he replied that “there is no need because we show the emails”.

However, this morning, when PO published another advert using a similar technique, it wrote in the original social media post that the “voiceover is generated by AI”. The video itself, though, did not contain such a warning.

Notes from Poland is run by a small editorial team and published by an independent, non-profit foundation that is funded through donations from our readers. We cannot do what we do without your support.

Main image credit: Platforma Obywatelska/Twitter (screenshot)

Daniel Tilles

Daniel Tilles is editor-in-chief of Notes from Poland. He has written on Polish affairs for a wide range of publications, including Foreign Policy, POLITICO Europe, EUobserver and Dziennik Gazeta Prawna.