AI-Generated War Propaganda on Music Platforms
AI-Generated War Propaganda on Music Platforms
It’s surprisingly easy to distribute AI-generated war propaganda on music platforms like SoundCloud and YouTube. 🎶 It’s cheap and easy to generate war propaganda, and it’s just as simple to distribute it on major music streaming platforms that do little to prevent it, according to Ukrainian intelligence researchers.
Terrecon, a Kyiv-based intelligence NGO, has found hundreds of AI-generated songs containing Russian war propaganda across major streaming platforms, including Deezer, Spotify, SoundCloud, and YouTube. The analysis identified 981 tracks showing signs of AI generation and Russian military propaganda. Of the 608 tracks with publicly available play counts, the combined total was almost 87 million plays. Most of the tracks identified by researchers had relatively low play counts, but 25 tracks had more than 1 million plays, and the most-played track had over 4.7 million plays. Many contained strong language and were not marked as AI-generated, except on Deezer, which marks AI content as such.
Researchers also found that propaganda songs can be produced quickly using AI music-generation tools like Suno or Udio and then easily distributed on major streaming platforms with little or no moderation. “We uploaded test tracks with explicit hate speech to YouTube, Spotify, Deezer, and SoundCloud. And nothing happened,” says Daria Verbytska, CEO and co-founder of Terrecon.
To see how quickly such tracks can be produced and how much they cost, Terrecon contacted 25 AI music creators advertising their services online. Of those, 15 agreed to produce a track, which cost anywhere between $11 and $330, and took as fast as 2 hours to 3 days to produce.
Terrecon also tested how streaming platforms respond to AI-generated war propaganda. Its analysts used Udio and Suno to produce 2 Russian-language songs containing profanity, glorifying the war, and calling for violence against Ukrainians. Udio, an AI music model, generated the track “without any restrictions,” according to Terrecon. Researchers said they used “a direct prompt in Russian that made no attempt to conceal the track’s propagandistic purpose,” including references to the “special military operation,” which is how Moscow calls its war against Ukraine, and “destroying enemies.”
Meanwhile, a model from Suno changed specific words or phrases but kept “the overall meaning and intent” of the prompt. The researchers then uploaded the tracks on YouTube and SoundCloud. They explicitly identified their purpose and included references to Russian war propaganda in their descriptions. The test found that YouTube did not automatically detect AI-generated audio and relied on creators to disclose the use of the technology in visual content. SoundCloud didn’t flag or label the content either. Neither YouTube nor SoundCloud removed the tracks immediately, and they stayed on the platforms for weeks.
Terrecon identified 15 main narratives in over 900 songs it analyzed, with most targeting the general public in Russia, but some also children and Ukrainians, including people in occupied territories. The glorification of the Russian military was the most common narrative in the lyrics, with other frequent themes referring to the “special military operation,” the mobilization, and the so-called “Russian world.” Music can be a more effective way to spread propaganda than traditional or social media because “the melody gets stuck in your head,” according to Verbytska. “Music bypasses the rational filter. When someone reads a propaganda article, there’s a chance critical thinking kicks in. When they listen to a song in the background, there isn’t,” she says.
“Removing a track means losing money. So moderation systems are built to catch specific banned words in a query, not to analyze overall meaning and context,” says Terrecon CEO Darya Verbytska. While Russian propaganda was the focus of research, the technology can be exploited by other threat actors. “These technologies are available to everyone. What is being tested against Ukraine today will be used against any democracy holding elections tomorrow,” Verbytska says, adding that platforms and regulators must do more to prevent AI-generated influence campaigns.