Spotify to label AI music, filter spam and more in AI policy change

Spotify on Thursday announced a series of updates to its AI policy, designed to better indicate when AI is being used to make music, to cut down on spam, and to make it clearer that unauthorized voice clones are not permitted on its service.

The company says it will adopt an upcoming industry standard for identifying and labeling AI music in credits, known as DDEX, and will soon roll out a new music spam filter to catch more bad actors.

Under the DDEX system, labels, distributors, and music partners submit standardized AI disclosures in music credits. This solution offers detailed information about the use of AI — like whether it was used for AI-generated vocals, instrumentation, or post-production, for example.

“We know the use of AI is going to be a spectrum, with artists and producers incorporating AI in various parts of their creative workflow,” said Sam Duboff, Spotify’s Global Head of Marketing and Policy, in a press briefing on Wednesday. “This industry standard will allow for more accurate, nuanced disclosures. It won’t force tracks into a false binary where a song either has to be categorically AI or not AI at all,” he noted.

As part of the same announcement, Spotify clarified its polices around AI-enabled personalization, stating directly thatunauthorized AI voice clones, deepfakes, and any other form of vocal replicas or impersonation are not allowed and will be removed from the platform.

While the DDEX standard is developing, Spotify says it’s received commitments from 15 labels and distributors who plan to adopt the technology, and sees its move as one that could signal to others it’s time to adopt the technology.

Because AI tools make it easier for anyone to release music, Spotify also has a new plan to cut down on the potential spam that results. This fall, the company will roll out a new music spam filter that will attempt to address spam tactics, tag them, and then stop recommending those tracks to users.

“We know AI has made it easier than ever for bad actors to mass upload content, create duplicates, use SEO tricks to manipulate search or recommendation systems…we’ve been fighting these kinds of tactics for years,” Duboff said. “But AI is accelerating these issues with more sophistication, and we know that requires new types of mitigations.”

The company said it would roll out the filter gradually to make sure it’s targeting the right signals, then add more signals over time as the market evolves.

Related to this, Spotify will also work with distributors to address something called “profile mismatches,” a scheme where someone fraudulently uploads music to another artist’s profile across streaming services. The company said it hopes to prevent more of these before the music ever goes live.

Despite the changes, Spotify executives emphasized that they still support use of AI provided it’s used in a non-fraudulent way. “We’re not here to punish artists for using AI authentically and responsibly. We hope that artists’ use of AI production tools will enable them to be more creative than ever,” noted Spotify VP and Global Head of Music, Charlie Hellman. “But we are here to stop the bad actors who are gaming the system, and we can only benefit from all that good side of AI if we aggressively protect against the downside,” he said.

Spotify’s updates follow a rapid increase in AI-generated music across the industry. This summer, an AI-generated band called Velvet Sundown went viral on its service, leading users to complain that the company isn’t transparent about labeling its AI tracks. Meanwhile, streaming rival Deezer recently shared that about 18% of the music uploaded each day to its service — or more than 20,000 tracks — is now fully AI-generated. (Deezer tells us that figure has now increased to over 30,000).

Spotify wouldn’t share its own metrics on the matter directly — but Duboff told reporters that “the reality is, all streaming services have almost exactly the same catalog.”

“People tend to deliver the music to all services,” he explained, adding that uploading tracks doesn’t mean anyone’s listening or that the AI music makes money. “We know AI usage is increasingly not a binary, but kind of a spectrum of how artists and producers are using it.”