Streaming music giant Spotify released a series of AI policy updates on Thursday, aimed at better identifying AI-generated music, reducing spam content, and clearly prohibiting unauthorized voice cloning on its platform.
The company announced it will adopt the industry standard DDEX to identify and label the use of AI in music works, and is about to launch a new music spam filter to combat more malicious actors.
Under the DDEX system, record labels, distributors, and music partners must submit standardized AI disclosure information in the music work details. This solution provides detailed information about AI usage, such as whether it is used for AI-generated vocals, instrumental performances, or post-production.
Sam Duboff, Spotify's global marketing and policy director, said at a media event on Tuesday: "We know that the use of AI will be on a spectrum, and artists and producers may integrate AI technology at various stages of the creative process. This industry standard will support more accurate and detailed disclosure methods, without forcing music works to be simply categorized as AI or non-AI."
As part of the same announcement, Spotify clarified its policies regarding AI personalization services, directly stating that unauthorized AI voice cloning, deepfakes, and any other form of sound replication or imitation are not allowed and will be removed from the platform.
Although the DDEX standard is still under development, Spotify said it has received commitments from 15 record labels and distributors to adopt the technology, believing this move could signal the industry to adopt the technology.
Due to AI tools making it easier for anyone to publish music, Spotify has developed a new plan to reduce potential spam content. In the fall this year, the company will launch a new music spam filter, attempting to identify spam tactics, mark them, and then stop recommending these tracks to users.
Duboff said: "We know AI makes it easier for bad actors to upload content in bulk, create duplicate content, and use SEO techniques to manipulate search or recommendation systems... We have been fighting against these strategies for years. But AI is accelerating these issues in more complex ways, and we know this requires new mitigation measures."
The company said it will gradually roll out the filters to ensure they target the right signals, and then add more signals as the market evolves.
In relation to this, Spotify will also work with distributors to address the so-called "catalog mismatch" issue, which involves someone fraudulently uploading music to the catalog of other artists on various streaming services. The company said it hopes to prevent more such behavior before the music is released.
Despite these changes, Spotify executives emphasized that they still support the non-fraudulent use of AI. Charlie Hellman, Vice President and Global Music Head at Spotify, stated: "We are not trying to punish real and responsible users of AI. We want artists to use AI creation tools to be more creative than ever before. But we need to stop those who cheat the system. Only by actively preventing the negative impacts of AI can we benefit from all the positive aspects of AI."
Spotify's updates come amid the rapid growth of AI-generated music across the industry. This summer, an AI-generated band called Velvet Sundown gained popularity on its service, leading users to complain about the lack of transparency in marking AI tracks. At the same time, streaming competitor Deezer recently revealed that approximately 18% (more than 20,000 tracks) of the music uploaded to its service each day is now entirely AI-generated. (Deezer told us this number has now increased to over 30,000 tracks).
Spotify does not directly share its metrics on this issue, but Duboff told reporters, "The reality is that all streaming services have almost identical music catalogs. People tend to upload music to all services," he explained. "Uploading a track does not mean someone is listening to it or that AI music can make money. We know that the use of AI is becoming less binary and is more of a spectrum of how artists and producers use it."
