Spotify Tightens AI Policy And Trims Catalog

Spotify announced on Thursday that it has culled 75 million tracks from its catalog that are “spammy,” as part of an effort to clamp down on the increasing flood of audio tracks submitted to the service that are vehicles for fraud, gaming the royalty system, or “AI slop.” It also announced a list of new policies that prohibit impersonation, spam submissions, and tracks used to generate fraudulent royalties. In this announcement, Spotify has joined Deezer in taking public steps to curb abuses in the huge and growing volume of music tracks submitted to it every day.

At the heart of this announcement is the unfathomably large and accelerating volume of music submitted to streaming music services every day. Deezer has estimated that the number is 150,000 – and that 28% of those are purely AI-generated. Music data firm Luminate tracks consumption data on over 200 million music tracks worldwide.

In removing 75 million tracks, Spotify has trimmed its catalog by a very significant percentage.

This explosion in volume – three orders of magnitude larger than during the CD era of the 1980s-1990s – is something that no one in the music industry has had any incentive to talk about until recently. Instead, it has been in the industry’s collective interest to maintain an image that anyone can record music, submit it to digital services, and with the right combination of talent and hard work, become the next Taylor Swift, Beyonce, or Bad Bunny.

But the incentives have been changing. It started with the major record labels, which are seeing diminishing market share of the catalogs available on streaming services. Universal Music Group CEO Sir Lucian Grainge complained in 2023 about the deluge of “functional, lower-quality content” that was diluting the royalty pool for human artists; these included tracks that barely exceeded the 30-second threshold for royalty eligibility. (Spotify recently increased the royalty eligibility threshold to two minutes for “functional noise recordings.”)

More recently, the incentives have been spreading to digital music services like Spotify and Deezer. Now that widely-available generative AI tools are enabling even further acceleration of track submission volume, the streaming services are finding reasons to crack down. Deezer claims the ability to detect fully AI-generated tracks with high accuracy; it states that 70% of these tracks are in some way fraudulent (e.g., attributed to fake artists).

Streaming music services are motivated to combat streaming fraud for various reasons: it reduces the amount of royalties that they pay legitimate labels and artists, and it harms users’ music discovery by distorting algorithms used to generate playlists, recommendations, and so on. Streaming fraud could also carry legal risks now that law enforcement is getting involved in prosecuting it. And at a practical level, the kinds of “spammy” tracks that Spotify has deleted from its library generate overhead for streaming services, taking up storage space and computing resources to manage it.

It has been inevitable that the complexity and scale of today’s music streaming services would lead to various ways to game the system. AI is an accelerant for those tactics, and not just through generative AI engines that can produce huge amounts of music tracks instantaneously. AI can also instantly generate the kinds of fake metadata, such as about artists and rights ownership, and massive amounts of spam tracks, that are also ingredients in system-gaming and fraud.

Questions remain about where the responsibility lies in the music supply chain for curbing streaming fraud and other abusive uses of AI. The non-profit Music Fights Fraud Alliance, which was formed in June 2023, says that it “unite[s] stakeholders across the industry in a coordinated fight against fraud.” Its membership consists of digital music distributors such as TuneCore, DistroKid, CD Baby, and Symphonic, and streaming music services such as Amazon Music, SoundCloud, Spotify, and YouTube Music.

Digital distributors also face issues of overhead as well as potential legal liability from these tactics. Digital distribution is a highly competitive market: there are dozens of digital distributors worldwide, many of which are generalists while some specialize in markets such as electronic or Latin music. They all place independent artists’ music tracks on dozens of digital music services around the world. They face pressures to accept as large a volume of tracks as possible to maximize revenue but also to appear as good actors that help promote music by true independent artists – while also competing on price against all the other digital distributors.

The leading digital distributors may well make their own announcements regarding AI, spam, and fraud in the coming months, which will be crucial in the development of that market.

Spotify, meanwhile, regardless of its new policies, is not banning music from its platform that is generated with AI tools. On the contrary, it’s contributing to standards being developed by the music industry standards body DDEX for identifying which parts of a music track were generated by AI. This information is intended to be captured during music creation and production, and carried down the supply chain to services like Spotify and Deezer to listeners. Spotify says that it intends to use that information for display to users; it hasn’t said anything about using that information in royalty calculations or other ways.

Spotify has also not announced any actions to deprivilege pure AI-generated tracks – unlike Deezer, which has technology for detecting them, does not include them in discovery algorithms, and does not pay royalties on them at all.

The other major digital music services – Apple Music, YouTube Music, and many others worldwide – are undoubtedly grappling with this same set of issues as the glut of released music tracks continues to accelerate. We may also see announcements from other services in the same vein as Spotify’s and Deezer’s. If the digital distributors and music services are successful in curbing abuses, then it will be interesting to see how much music humans actually generate with the help of AI tools. The tools have great potential for creativity as well as volume.

BillboardLucian Grainge Calls For ‘Updated Model’ For Music Industry: Read His Memo to UMG Staff