AI rights are becoming product-specific.
Fan remixing, adaptive gaming, model training, and generated lyrics do not require one blanket “AI right.” They require different permissions, evidence, and ways of getting paid.
For much of the music industry’s early AI debate, “AI rights” sounded like a single question: did a company have permission to train a model on protected music?
That question still matters. But this week’s developments show that the commercial reality is becoming more specific—and more complicated.
A fan remix on Spotify, a song that adapts to gameplay, a model trained on lyrics, and an output that reproduces protected expression are not the same use. They involve different assets, permissions, participants, risks, and ways of getting paid.
For independent artists and publishers, the next generation of agreements cannot rely on one broad AI clause that tries to cover everything. Rights need to be defined at the level of the product being built.
LICENSING AGREEMENT / FORTHCOMING PRODUCT
Spotify is turning licensed fan transformation into a product.
On October 20, Spotify and Believe announced an agreement that would allow participating Believe and TuneCore artists and labels to opt into Spotify’s forthcoming AI-powered fan covers and remixes tool. Spotify says the feature is planned as a paid Premium add-on and that streams of participating works will generate royalties for artists and rights holders. Believe says participation will require two separate opt-ins.
This is not a launched product yet, and the announcement does not answer every important commercial question. Still, it marks a meaningful shift. AI-enabled fan creation is being framed not merely as user activity to tolerate or police, but as a licensed experience from which artists and rights holders may earn revenue.
An opt-in should say more than “yes to AI.” It should identify the product, the assets, and the rules.
What needs to be cleared
A fan-made cover or remix can implicate the composition, the master recording, vocals, artist identity, adaptation rights, and rules governing the resulting version. A useful agreement should identify what is authorized, how a transformed work will be credited, how revenue will be divided, whether participation can be withdrawn, and what happens to versions created before withdrawal.
Worth reading: Spotify’s official partnership announcement ↗
LICENSING / BUSINESS-MODEL DEVELOPMENT
Adaptive gaming is becoming a new sync-adjacent license.
Two days later, Universal Music Group and TYGER announced a strategic licensing agreement covering selected recorded-music and publishing catalogs. TYGER’s technology is designed to adapt licensed songs to gameplay in real time. Instead of placing one fixed recording in one fixed scene, the system can reshape the listening experience in response to what happens during play.
TYGER says its existing catalog includes more than 400,000 songs, primarily from independent artists, and that its service is being developed with subscription and ad-supported tiers.
The more dynamic the product, the more precise the permission
This creates potential licensing value for independent catalogs, but it also raises questions conventional sync language may not answer. How much may a track change? Which elements may be separated or recombined? Are composition and master approvals aligned? Which games or gameplay contexts are permitted? How will uses be measured, attributed, and paid?
Worth reading: UMG and TYGER’s official licensing announcement ↗
LITIGATION MILESTONE / NOT A RULING
The Anthropic lyrics dispute reached the fair-use question.
On October 21, U.S. District Judge Eumi Lee heard arguments over whether Anthropic’s alleged use of copyrighted lyrics to train Claude qualifies as fair use.
According to Reuters, music publishers argued that Claude can generate lyrics that compete with protected works and pointed to nearly 120,000 requests for synthetic lyrics during a six-month period. Anthropic argued that training is transformative and that the publishers had not shown specific market harm. The judge did not issue a ruling at the hearing.
That distinction matters. This was a litigation milestone, not a finding that Anthropic infringed copyrights or that its conduct was fair use.
Market evidence is becoming part of rights management
Rights holders may benefit from documenting actual or emerging licensing markets, requests for synthetic substitutes, reproductions of protected expression, and the ways generated material competes with human-written lyrics. A clean catalog and registration record remain essential, but so may evidence of how a work is licensed, where substitutes appear, and how those substitutes affect value.
Worth reading: Reuters’ report on the hearing ↗
NEW COPYRIGHT & DMCA LITIGATION / ALLEGATIONS
Reservoir’s lawsuits show why provenance and copyright information matter.
Also on October 21, independent publisher Reservoir filed separate complaints against Suno and Anthropic. Reservoir alleges that Suno copied 1,050 compositions it exclusively controls and that Anthropic copied lyrics from 853 compositions. The Suno complaint also includes claims involving circumvention of access controls and the removal or alteration of copyright-management information.
These are allegations, not court findings. But the complaints are significant because they extend the dispute beyond the abstract question of whether model training can be fair use. They also focus on how works were allegedly obtained, whether ownership information was preserved, and whether the systems produced protected material.
Metadata and provenance are operational assets
A publisher’s catalog record should establish which rights it controls, whether that control is exclusive, who has authority to license or enforce those rights, what copyright-management information was attached to the work, and what evidence exists when that information disappears or protected expression resurfaces.
Worth reading: Reservoir’s complaint against Suno ↗, its complaint against Anthropic ↗, and Music Business Worldwide’s summary ↗
THE PLAYFUL BEAR TAKEAWAY
Manage the use—not just the label.
“AI rights” is no longer a useful description unless it is followed by a specific use. Fan remixing, adaptive gaming, model training, generated lyrics, synthetic vocals, and artist-identity experiences each require their own permissions and economic terms.
At Playful Bear Music, we want PB RightsHub to make those distinctions visible. As we develop that infrastructure, our goal is to build a product-level rights record for every AI-enabled opportunity involving music we represent.
- 01Define the product and the permission
Identify the proposed use, the allowed transformations, and the limits that apply to training, generation, distribution, or audience participation.
- 02Map every asset and authority
Connect the composition, master, vocal, identity, and other protected elements to the people or organizations authorized to approve them.
- 03Record the economics and the exit
Track compensation, reporting, attribution, duration, territory, withdrawal rights, and the treatment of outputs that already exist.
- 04Preserve provenance and evidence
Retain model information, source-material representations, copyright-management information, approvals, and the evidence needed for licensing or enforcement.
This product-level record should sit alongside—not replace—our work on Human Creative Authority. HCA-6 helps document meaningful human contribution across Intent, Direction, Selection, Assembly, Interpretation, and Finishing. It does not, by itself, establish whether a model’s training material was lawfully acquired or whether every third-party right has been cleared.
Keeping those layers separate is how Playful Bear Music and PB RightsHub can avoid turning “human-created” or “licensed AI” into vague marketing claims.
THE BOTTOM LINE
Precision is becoming creator leverage.
The market is moving from a debate about whether AI belongs in music to a negotiation over which AI-enabled uses belong in which agreements.
The practical move is not to write a longer blanket AI clause. It is to build a clearer permissions system. Before entering an AI-related partnership, identify the exact product, assets, uses, outputs, and revenue flows. Preserve the chain of title and copyright-management information. Record what each participant approved. Ask how withdrawal works and what survives it. Keep the authorship record distinct from the provenance and licensing record.
The opportunity is real. So is the administrative burden. Independent music businesses that can describe their rights clearly—and prove what they control—will be in the strongest position to participate on their own terms.
About this series
The Music Rights Brief is a weekly field note from Playful Bear Music, an imprint of The Publishing Division. PB RightsHub is The Publishing Division’s developing rights-information infrastructure. The series tracks legal, licensing, platform, and business developments affecting independent music. It is educational analysis, not legal advice.