What are streaming services going to do about the explosion of fully AI-generated music being uploaded to their platforms? It\u2019s a big question in 2025, and so far it has mainly been Deezer answering it publicly.<\/p>\n\n\n\n
The company has revealed the scale of AI music on its service \u2013\u00a028% of uploads but only 0.5% of streams<\/a> \u2013\u00a0and has a policy of clearly labelling full-AI tracks, and not including them in its editorial playlists or algorithmic recommendations.<\/p>\n\n\n\n Now the bigger streaming players are starting to set out their policies. Including Spotify, which has today announced three measures to respond to the rise of GenAI music.<\/p>\n\n\n\n\n They include supporting a new industry standard for music credits, developed by the DDEX consortium, to make it clear how and when AI was used in music tracks.<\/p>\n\n\n\n Spotify is also rolling out a new spam filter that it says will identify \u2018bad actors\u2019 mass-uploading tracks for fraudulent purposes and\/or trying to manipulate its recommendation systems.<\/p>\n\n\n\n Finally, the company is updating its policies to clarify that unauthorised AI voice-clones, deepfakes and other \u201cvocal replicas and impersonation\u201d are not allowed on Spotify \u2013\u00a0and will be removed if detected.<\/p>\n\n\n\n\n The three measures were announced by Spotify during an online press event, with VP, global head of music vertical Charlie Hellman and global head of marketing and policy at Spotify for Artists Sam Duboff. Read on for the details.<\/p>\n\n\n The DDEX standard for disclosing use of AI in music credits hadn\u2019t been officially announced before, but Spotify is revealing its support ahead of its launch in an attempt to build momentum behind it within the music industry.<\/p>\n\n\n\n \u201cIf a track includes AI-generated vocals, AI instrumentation, a specific instrument that\u2019s AI-generated, or AI-assisted post-production like mixing or mastering, that information can be displayed in the credits on Spotify or any other service using the standard,\u201d said Duboff.<\/p>\n\n\n\n\n \u201cAs content providers begin to submit the metadata to us, over time, we\u2019ll be displaying it across Spotify.\u201d<\/p>\n\n\n\n Duboff said that Spotify\u2019s adoption of the DDEX standard is an attempt to take a more nuanced approach to which tracks should be labelled and how.<\/p>\n\n\n\n \u201cWe know the use of AI is going to be a spectrum, with artists and producers incorporating AI in various parts of their creative workflow, and this industry standard will allow for more accurate, nuanced disclosures,\u201d he said.<\/p>\n\n\n\n \u201cIt won\u2019t force tracks into a false binary where a song either has to be categorically AI or not AI at all\u2026 Listeners get clarity, and artists get a consistent way to share their process with listeners across all services, without fear that it\u2019ll affect how their work\u2019s promoted.\u201d<\/p>\n\n\n\n This only works if labels and distributors use the new standard, of course. Spotify says that 15 have already signed up, and has named them:<\/p>\n\n\n\n Amuse, AudioSalad, Believe, CD Baby, DistroKid, Downtown Artist & Label Services, Empire, EMS GmbH, FUGA, Idol, Kontor New Media, Labelcamp, NueMeta, Revelator, SonoSuite, Soundrop and Supply Chain are the early adopters.<\/p>\n\n\n\n Spotify is very keen to stress that the new credits system isn\u2019t about naming and shaming artists though.<\/p>\n\n\n\n \u201cWe\u2019re not here to punish artists for using AI authentically and responsibly,\u201d said Hellman. \u201cWe hope that artists use of AI production tools will enable them to be more creative than ever, create more content that excites fans and offer the best possible experience on Spotify.\u201d<\/p>\n\n\n\n Stamping out music fraud \u2013 which often relies on GenAI tools to create the sheer scale of tracks required for scams to pay off \u2013\u00a0is a big priority for Spotify and its rivals.<\/p>\n\n\n\n Apple Music made a rare industry-conference speech<\/a> at our Music Ally Connect event this year to talk about its fraudbusting work, and Spotify\u2019s new spam filter builds on the latter\u2019s existing efforts to tackle these schemes.<\/p>\n\n\n\n \u201cOver the last year, we\u2019ve actually removed more than 75 million spammy tracks from Spotify,\u201d said Hellman. But this is a game of whacking, well, tens of millions of moles.<\/p>\n\n\n\n \u201cAI has made it easier than ever for bad actors to mass upload content, create duplicates, use SEO tricks to manipulate search or recommendation systems,\u201d was how Duboff described the challenge.<\/p>\n\n\n\n \u201cWe\u2019ve been fighting these kinds of tactics for years, but AI is accelerating these issues with more sophistication, and we know that requires new types of mitigations. Our new spam filter will flag tracks and uploaders using these tactics and stop recommending them across Spotify programming.\u201d<\/p>\n\n\n\n The filter is rolling out gradually, which Spotify says is to ensure that it is targeting the right content.<\/p>\n\n\n\n \u201cWe\u2019ll keep adding new detection signals as we learn more. This matters, because spam and slop can dilute the royalty pool and clog up the system, hurting professional artists and songwriters who are following the rules,\u201d said Duboff.<\/p>\n\n\n\n \u201cCutting it out means more visibility, more revenue for those following the rules, and a better experience for listeners, too.\u201d<\/p>\n\n\n\n He also outlined exactly what Spotify means when it talks about \u2018spam\u2019 or \u2018slop\u2019 on its service.<\/p>\n\n\n\n \u201cWe\u2019re looking for patterns like mass uploads from the same account; excessive duplicates of similar songs with slightly different metadata; intentional SEO manipulation and hacking; artificially-short-track abuse, where you tend to upload songs just slightly longer than 30 Seconds to try to rack up royalty-bearing streams,\u201d said Duboff.<\/p>\n\n\n The third measure announced today by Spotify is the policies clarification about voice clones and deepfakes, simplifying and speeding up the process of removing them when notified by label and artist teams.<\/p>\n\n\n\n This will not affect artists who have chosen to create and license their voices out with AI partners, however \u2013\u00a0a task that a growing number of startups are popping up to handle.<\/p>\n\n\n\n \u201cIf they [artists] want to license their voice to an AI project, that\u2019s their choice. Our job is to make sure it stays their choice,\u201d said Duboff.<\/p>\n\n\n\n This is part of wider efforts that go beyond AI to other forms of impersonation, he added. That includes a problem that has been in the news this summer: music being uploaded to established artists\u2019 Spotify profiles that they didn\u2019t make.<\/p>\n\n\n\n \u201cWe are also increasing our investments in stopping what we call \u2018profile mismatches\u2019. We\u2019ve got a lot of feedback from artists about this,\u201d said Duboff.<\/p>\n\n\n\n \u201cWe\u2019re excited to be working directly with distributors on some new tests to block these attacks before they even get to streaming streaming platforms.\u201d<\/p>\n\n\n During the announcement event, Hellman and Duboff talked about the wider issues around AI, including fielding journalists\u2019 questions about today\u2019s news.<\/p>\n\n\n\n \u201cSome people are really excited about AI: they see new, creative doors opening. And others are sceptical: they see the risks. They\u2019re worried about protecting what\u2019s already working. And honestly, both reactions make total sense to me,\u201d said Hellman.<\/p>\n\n\n\n He compared the current discourse around AI to past discussions around multi-track tape machines, synthesizers, DAWs and AutoTune, in terms of there being split opinions within the artist community about their value.<\/p>\n\n\n\n \u201cThe common thread in each of those moments was that it was the artist that got to decide, and the job of the rest of us in the industry was to make sure that those decisions stayed in their hands,\u201d he said.<\/p>\n\n\n\n However, there are differences this time round: the speed and scale at which AI is growing; the impact that is having on artists; and the questions it is posing for companies like Spotify.<\/p>\n\n\n\n \u201cAt its best, AI can help artists experiment and finish ideas faster, and find new audiences,\u201d said Hellman.<\/p>\n\n\n\n \u201cBut we also know that at its worst, it can be used to impersonate someone\u2019s voice without their permission; flood the system with low value content \u2013 the kind of slop that nobody\u2019s asking for \u2013 and it can make it harder for real artists to cut through.\u201d<\/p>\n\n\n\n Spotify is treading carefully so as not to be accused of overreach by artists who are using AI as part of their creative process.<\/p>\n\n\n\n \u201cThis isn\u2019t about gatekeeping, it\u2019s about building trust across the whole music ecosystem,\u201d said Duboff. Both execs also stressed that today\u2019s moves are part of a wider plan to respond and adapt to GenAI developments.<\/p>\n\n\n\n \u201cThis won\u2019t be the end of the story. AI is evolving fast, but today\u2019s announcements will be a meaningful step forward,\u201d said Hellman.<\/p>\n\n\n\n \u201cWe\u2019re going to keep talking to artists, songwriters, producers, distributors, about what\u2019s working and what\u2019s not, and when new changes come up \u2013 and I\u2019m sure that they will \u2013 we\u2019ll act.\u201d<\/p>\n\n\n\n \u201cThis is the start of a longer journey. We\u2019ll keep learning and adjusting, rolling out updates and making the music ecosystem more transparent and more fair for artists, for rightsholders and for the millions of listeners who trust Spotify to deliver the real thing.\u201d<\/p>\n\n\n\n During the Q&A, Music Ally asked if Spotify could share any stats comparable to Deezer\u2019s announcements about fully AI-generated music having grown from 10% of its uploads in January to 28% now \u2013\u00a0while still only accounting for 0.5% of streams.<\/p>\n\n\n\n In his answer, Duboff steered clear of specifics, but said that Spotify sees a \u201cde minimis level of engagement\u201d from listeners in these tracks. \u2018De minimis\u2019 being a legal term meaning \u2018trifling\u2019 or \u2018too minor to merit consideration\u2019 if you\u2019re wondering.<\/p>\n\n\n\n \u201cA really, really small percentage of streams,\u201d he added. \u201cIn general, when the music doesn\u2019t take much effort to create, it tends to be low quality and doesn\u2019t tend to find an audience.\u201d<\/p>\n\n\n\n Hellman returned to the point about the wider spectrum of how AI is used for music, which stretches far beyond what he described as \u201cprompt-generated music\u201d.<\/p>\n\n\n\n \u201cInitially, I think people\u2019s mindset was very much binary. There\u2019s either AI music or there\u2019s not. But the reality is that we\u2019re now seeing this proliferation of so many different ways that AI is incorporated into all different steps of the tool-chain when people are making music,\u201d he said.<\/p>\n\n\n\n He cited some of the lyrics for KPop Demon Hunters being written as a \u201ccollaboration\u201d with ChatGPT; decades-old songs reworked with the help of AI stem-separation technology; AI\u2019s use to translate vocals into different languages and more.<\/p>\n\n\n\n \u201cFor us, when we think about our policies, we only react to music once it\u2019s delivered to Spotify, and then we want to make sure that the tactics that are being used to promote it and gain streams are fair and not gaming the system or deceptive,\u201d he said.<\/p>\n\n\n Deezer made a big announcement earlier this year that it had developed a tool to detect fully AI-generated tracks. It seems Spotify is cautious about that approach, even while exploring it internally.<\/p>\n\n\n\n \u201cWe do engage in some detection software so that we can get aggregate stats. They\u2019re imperfect, and there\u2019s a lot of false positives, but there are ways that we can kind of test the waters overall, to get a general sense of the amount that seems to be purely prompt-generated.\u201d<\/p>\n\n\n\n \u201cThat\u2019s helpful to see trends, but it\u2019s really not helpful for policy enforcement, again, because this isn\u2019t binary, and because there\u2019s so many false positives, and none of those detection systems are foolproof.\u201d<\/p>\n\n\n\n Spotify is also clearly keen not to be seen as going it alone on some of this work. For example its adoption of the DDEX standard rather than rolling out its own labelling system, as Deezer did.<\/p>\n\n\n\n \u201cWe think the industry is better off and listeners are better off with industry standards,\u201d said Duboff.<\/p>\n\n\n\n \u201cIf we think about a world where every streaming service is building their own methodology, their own technology, their own system for what they would call AI and what they don\u2019t, [and where] every label and distributor would have their own methodology and language they\u2019d use to try to deliver music and give AI information, that doesn\u2019t seem like how we build listener trust as an ecosystem.\u201d<\/p>\n\n\n\n \u201cImagine if you go to Spotify to listen to a song, it has one piece of AI information, you go to another service, and there\u2019s a whole different set of information. You go to another and there\u2019s nothing at all. That doesn\u2019t seem like a way to really preserve that kind of authentic belief in music and artistry that we know is so important for fans to maintain.\u201d<\/p>\n\n\n\n This could be seen as a not-so-implicit criticism of Deezer\u2019s approach, but this is a very fluid area. There\u2019s no reason why Deezer couldn\u2019t also adopt the DDEX system, and collaborate with Spotify and other streaming services on detection tools and other measures.<\/p>\n\n\n\n He also addressed one of the big AI-music stories of this summer: the rapid rise to prominence of GenAI band The Velvet Sundown<\/a>, who at one point had 1.5 million monthly listeners on Spotify \u2013\u00a0although that total has slumped to less than 300,000 now.<\/p>\n\n\n\n \u201cWe definitely were interested in all the commentary on that one. It generated a lot of conversation, and a lot of cases like that have helped inform the approach we\u2019ve taken,\u201d he said.<\/p>\n\n\n\n \u201cWe don\u2019t think it\u2019s our role to dictate or police how an artist chooses to present their work. The top 50 charts are littered with artists who perform under pen names and stage names and band names. And we don\u2019t police how an artist wants to express themselves: how they want to name their artist project,\u201d continued Duboff.<\/p>\n\n\n\n Instead, Spotify is focusing on the three themes reflected in today\u2019s announcement: spam, impersonation and deception. He also suggested that if the new DDEX credits system had been in place earlier this year, The Velvet Sundown may have provoked less controversy.<\/p>\n\n\n\n \u201cRight now, across every streaming service, the only real method you have to disclose the provenance of your tracks is in freeform artist bios on your profile, which differ on each service and tend to be a little bit more buried,\u201d said Duboff.<\/p>\n\n\n\n \u201cIf these AI credits have been available before the summer. I think the news cycle, [and] the fan interest would have been really different. If you went to each track from a band like The Velvet Sundown, they were given the opportunity to clearly share: \u2018Here is the extent to which it was AI-generated; here ar some of the areas that was human input; here\u2019s who the humans were.\u2019\u201d<\/p>\n\n\n\n In other words, if the people behind a GenAI artist like The Velvet Sundown (or this month\u2019s big story, Xania Monet<\/a>) are being transparent about what they\u2019re doing, Spotify is comfortable with this music being on its service.<\/p>\n\n\n\n However, once the new credits roll out \u201cif someone decided not to use them, that\u2019s different,\u201d said Duboff. \u201cThat\u2019s the sort of framework we want to set up so we can evolve and think about what deception means in the age of AI.\u201d<\/p>\n\n\n One final question focused on Spotify\u2019s own role (or rather, its potential role) in making GenAI music for its service.<\/p>\n\n\n\nAI transparency in music credits<\/h2>\n\n\n
A crackdown on AI slop and music spam<\/h2>\n\n\n
Taking on the AI-music deepfakes<\/h2>\n\n\n
\u2018This isn\u2019t about gatekeeping\u2019<\/h2>\n\n\n
![\"tumbleweed](\"https:\/\/i0.wp.com\/musically.com\/wp-content\/uploads\/2025\/09\/ziul3lwgqim-e1758795555901-700x401.jpg?resize=700%2C401&ssl=1\")
<\/figure>\n\n\nAI music\u2019s \u2018de minimis\u2019 engagement<\/h2>\n\n\n
\u2018Better off with industry standards\u2019<\/h2>\n\n\n
![\"The](\"https:\/\/i0.wp.com\/musically.com\/wp-content\/uploads\/2025\/06\/Screenshot-2025-06-26-at-16.37.09.png?resize=700%2C336&ssl=1\")
<\/figure>\n\n\nAddressing the Velvet Sundown challenge<\/h2>\n\n\n
\u2018Spotify doesn\u2019t generate any music\u2019<\/h2>\n\n\n