By Lucas Paolini, MA, Events Coordinator, SELS – October 2025
This article unpacks the upcoming deals being struck by major music labels and AI platforms. What is expressed below is solely the opinion of SELS with reference to publicly available information as of October 5, 2025
The repercussions of artificial intelligence are no longer hypothetical in the music business—they’re being negotiated into contracts. According to the Financial Times, major labels Universal Music Group (UMG) and Warner Music Group (WMG) are within weeks of landmark licensing agreements with AI companies that would formalise how those firms pay to use music in training and to generate new works. The talks reportedly span start-ups like ElevenLabs, Stability AI, Suno, Udio and Klay Vision, as well as Big Tech platforms including Google and Spotify. While details are fluid, one central aim is to establish a “streaming-style” system so each qualifyingcan use generates a micropayment to rightsholders, supported by attribution technology akin to YouTube’s Content ID.
Sony Music is also in the mix. While not as close to the finish line as UMG and WMG, Sony has signalled its engagement with companies whose models are “ethically trained” and structured to benefit artists and songwriters, another sign that the major labels are seeking to channel AI’s momentum rather than merely fight it in court.
Why now? Because AI music is Already Everywhere
The urgency is obvious on the platforms. French streaming platform Deezer said in September that roughly 28% of tracks uploaded daily are now fully AI-generated, a staggering number that illustrates how easily synthetic audio can flood catalogues. Spotify, for its part, disclosed that it removed over 75 million “spammy” tracks in the past 12 months. While not all of those removals were necessarily AI-made, the timing aligns with the surge of generative audio tools and manipulation tactics. This isn’t the industry’s first technology shock. A generation ago, labels were routed by peer-to-peer sharing networks like Napster and LimeWire. One of the defining flashpoints was Metallica v. Napster (2000), a lawsuit that resulted in court injunctions forcing Napster to block unauthorized sharing of Metallica’s catalogue and contributed to the service’s shutdown and eventual bankruptcy proceedings as its assets were sold off. LimeWire was later ordered to shut down in 2010 after a federal ruling found it liable for inducing infringement. Those cases cemented a lesson: ignore the tech wave, and someone else will write the rules.
The labels appear determined not to repeat the early-2000s missteps. Rather than waiting for chaos to calcify, they’re setting terms before generative music becomes ungovernable. As part of the pending AI deals, the majors are reportedly pressing for attribution and detection systems—think Content ID for training and outputs—so that uses of protected recordings, compositions, or artist likenesses can be identified and monetised at scale. Unlike the Napster era, where infringement often meant one-to-one copies of existing recordings traded without payment, generative systems can use copyrighted works in training without permission (raising reproduction and derivative-work issues), and generate outputs that may imitate an artist’s voice or the protectable elements of a composition or recording, even if no literal sample is present.
Both raise legal and business questions. In 2024, the major labels (via the Recording Industry Association of America) filed first-of-their-kind lawsuits against Suno and Udio (AI generating music platforms) alleging mass infringement by training on protected sound recordings. Those cases remain active and are widely seen as bellwethers for how U.S. courts will treat music-generation models. Separately, music publishers sued Anthropic over AI regurgitation of song lyrics; a federal judge declined to issue a preliminary injunction in March 2025, but the underlying claims continue. These disputes are the legal backdrop to the current push for licensed solutions.
The Payment Model Labels Want
If the deals close, expect payment mechanics that borrow from streaming economics. Labels want micropayments triggered by specific uses like training, prompting, or the playback of AI-generated tracks that rely on licensed material or identifiable artist likenesses. To make that work, AI companies would need reliable attribution tech: systems that can detect when a model uses or evokes protected assets and apportion royalties to the right parties. Industry insiders often cite YouTube’s Content ID, which matches user uploads against a database of reference files as the conceptual template, though deploying an equally robust system for model training and synthetic outputs is a harder technical challenge.
Generative music complicates a second legal front: voice and likeness. Even when an AI-made track doesn’t copy a recording or composition, it can still impersonate a distinctive voice. Law here is evolving quickly. In the United States there is no single federal “right of publicity,” but states are looking to be the catalysts for the regulatory generative AI legal framework. Tennessee’s ELVIS Act, effective July 1, 2024, explicitly protects voice against unauthorized AI cloning, reflecting the state’s music-industry roots. In Washington, the NO FAKES Act has been re-introduced in Congress in 2025 to create a national cause of action for unauthorized digital replicas, though it remains under debate. Europe is going a different route: the EU AI Act imposes transparency and copyright-compliance obligations on general-purpose models, with key provisions for generative AI coming into force during 2025–2026. These regulatory moves are reshaping the incentives for licensing, compliance and product design.
It’s fair to ask whether AI could drive a collapse akin to the early-2000s. Historically, global recorded-music revenues fell by roughly half from the 1999 peak to the mid-2010s trough before streaming revived growth. Today, however, the commercial environment is different: streaming subscriptions are entrenched, the platform economy is regulated, and labels are moving early to set terms for AI. IFPI data show global recorded-music revenues grew 4.8% in 2024 to about $29.6 billion, the tenth consecutive year of expansion—hardly a collapsing market. That doesn’t mean AI is benign, but it does suggest the business is negotiating from a position of relative strength.
That said, there are material risks. Catalogue dilution—the flood of low-quality or sound-alike tracks—can make it harder for legitimate releases to be discovered. Fraud and manipulation (fake artists, botted streams, “noise” uploads) siphon royalties from real creators. Spotify’s 75-million “spammy” takedown figure underscores the scale of the hygiene problem that platforms and labels now face. Attribution and enforcement tech will matter at least as much as headline royalty rates.
What a Sensible AI–Music Settlement Looks Like
A sustainable settlement between labels and AI companies will likely contain four pillars:
Clear licensing lanes: Distinguish between training on recordings/compositions, output uses that reference or reproduce protected elements, and voice/likeness uses. Each lane should carry its own license terms and rates. (This mirrors how sync, master, and publishing are distinct in traditional deals.)
Attribution infrastructure: A Content-ID-style system that can (a) trace training usage to rightsholders where required by law, (b) detect recognizable protected elements in outputs, and (c) verify consent for voice and likeness replicas.
Micropayments and revenue integrity: Streaming-like accounting for each qualifying use, with strong anti-fraud controls to avoid the garbage-in/garbage-out dynamic we’ve already seen with synthetic noise uploads.
Artist choice and control: Opt-in frameworks for voice models and “style” training; easy takedown of deceptive replicas; and transparent disclosures to users when content is AI-generated or voice-cloned. (Elements of this approach are echoed in the EU AI Act and the U.S. NO FAKES proposal.)
For readers mapping the star system to the corporate acronyms: Taylor Swift releases via Republic Records (a UMG label); Coldplay records for Warner (Atlantic in the U.S., Parlophone in the U.K.); Kendrick Lamar has released through Top Dawg Entertainment with distribution via Interscope (UMG). Those connections help explain why the majors’ stance matters so much: deals struck at the group level set the framework that protects their marquee acts’ recordings, compositions, and their voices.
From Litigation to Licensing
The courts will still matter. The RIAA’s suits against Suno and Udio (and the publishers’ case against Anthropic) will influence how far fair-use and other defences can stretch in the AI context. But the current wave of licensing talks shows the industry’s preferred endgame: channel AI into licensed, accountable products that compensate creators and preserve trust with listeners. If the majors can lock in attribution and consistent micropayments now, they might avoid a repeat of the Napster era’s value destruction while still leaving space for the creative upsides of new tools. In other words, the question is no longer whether AI belongs in music. It’s how it belongs—and who gets paid when it does.
Works Cited
‘Music labels close to landmark AI licensing deals’ Financial Times (3 October 2025) https://www.ft.com/content/1a1ae15b-af1a-4daf-8d1f-4c8a7db77865 accessed 5 October 2025.
Reuters, ‘Music labels near AI licensing deals — FT’ (3 October 2025) https://www.reuters.com/world/us/music-labels-near-ai-licensing-deals-ft-2025-10-03/ accessed 5 October 2025.
Murray Stassen, ‘Spotify has deleted 75m+ “spammy tracks” – as it unveils new AI music policies’ Music Business Worldwide (25 September 2025) https://www.musicbusinessworldwide.com/spotify-has-deleted-75m-spammy-tracks-as-it-unveils-new-ai-music-policies/ accessed 5 October 2025.
The Guardian, ‘Spotify removes 75m spam tracks in past year as AI increases ability to make fake music’ (25 September 2025) https://www.theguardian.com/music/2025/sep/25/spotify-removes-75m-spam-tracks-past-year-ai-increases-ability-make-fake-music accessed 5 October 2025.
The Hollywood Reporter, ‘Spotify Removes 75 Million “Spammy” Songs, Cracks Down on AI Use by “Bad Actors”’ (25 September 2025) https://www.hollywoodreporter.com/business/business-news/spotify-new-ai-policies-spam-filter-enforcement-1236379926/ accessed 5 October 2025.
Rolling Stone, ‘Spotify Won’t Ban AI Music Under New Rules’ (25 September 2025) https://www.rollingstone.com/music/music-features/spotify-not-banning-ai-music-new-guidelines-1235434946/ accessed 5 October 2025.
Blake Brittain, ‘Music AI startups Suno and Udio slam record label lawsuits in court filings’ Reuters (1 August 2024) https://www.reuters.com/legal/litigation/music-ai-startups-suno-udio-slam-record-label-lawsuits-court-filings-2024-08-01/ accessed 5 October 2025.
Blake Brittain, ‘US judge declines to block Anthropic from using song lyrics in AI training’ Reuters (28 March 2025) https://www.reuters.com/legal/us-judge-declines-block-anthropic-using-song-lyrics-ai-training-2025-03-28/ accessed 5 October 2025.
Financial Times, ‘Music industry revenues slow as pandemic-era CD boom fades’ (19 March 2025) https://www.ft.com/content/05ea07dc-2fae-4616-93e1-746cc8ac4635 accessed 5 October 2025.
Reuters, ‘Music revenues rise again in 2024, boosted by streaming subscriptions, report shows’ (19 March 2025) https://www.reuters.com/business/media-telecom/music-revenues-rise-again-2024-boosted-by-streaming-subscriptions-report-shows-2025-03-19/ accessed 5 October 2025.
Image credited to FreePik
