Rethinking Music Licensing in the Age of Generative AI

15 Jan Rethinking Music Licensing in the Age of Generative AI

The debate around artificial intelligence and music licensing is often framed as a binary choice: protect creators or allow unfettered technological creativity. In reality, this framing is already outdated. Generative AI has fundamentally altered how music is created, distributed, and consumed, and with it, the assumptions that underpin existing licensing frameworks.

The numbers are staggering. According to Deezer, fully AI-generated music now accounts for 34% of new tracks uploaded to its platform daily, roughly 50,000 synthetic songs, with a 400% increase since January 2025 (Deezer/Ipsos, 2025). Critically, Deezer reports that approximately 70% of streams on AI-generated tracks were flagged as fraudulent, driven by bots rather than genuine listeners. In a survey commissioned by Deezer and conducted by Ipsos across eight countries, 97% of respondents could not distinguish between AI-generated tracks and human-made music when presented with blind listening tests (Deezer/Ipsos, 2025). This technological reality demands more than incremental policy adjustments; it requires a fundamental reimagining of how we compensate creative work.

The Principle That Must Not Change

Before examining new frameworks, one principle must be stated unequivocally: original artists, performers, and rights-holders deserve compensation. That principle should not be diluted by technological change, commercial pressure, or industry convenience.

Today, the most responsible AI music platforms acknowledge this by working to establish licensing agreements with major rights-holders. In November 2025, all three major labels (Universal Music Group, Sony Music Entertainment, and Warner Music Group, along with their respective publishing arms) struck individual licensing deals with AI music startup Klay Vision (Variety, 2025). Warner Music also signed agreements with Stability AI and reached settlements with Udio. Universal Music Group announced a settlement with Udio in October 2025 (Music Business Worldwide, 2025).

These developments represent a dramatic shift from litigation to collaboration. Less than eighteen months earlier, in June 2024, the Recording Industry Association of America (RIAA), on behalf of those same three major labels, had filed landmark copyright infringement lawsuits against AI music generators Suno and Udio. The complaints alleged that these companies had copied sound recordings “en masse” and ingested them into AI models without authorization, seeking damages of up to $150,000 per work infringed, potentially hundreds of millions of dollars in total liability (RIAA, 2024).

Why These Approaches Remain Transitional

However, these early licensing agreements and legal settlements are best understood as transitional arrangements, not permanent solutions. They represent pragmatic hedges during a period of structural uncertainty, attempts to apply existing frameworks to fundamentally new technology.

Bloomberg reported in June 2025 that the major labels were seeking not only license fees but also “a small amount” of equity in both Suno and Udio as part of settlement negotiations (Bloomberg, 2025). According to reports in October 2025, Suno was simultaneously raising over $250 million at a valuation of $2.45 billion, with annual recurring revenue above $100 million (Level Law, 2025). These equity stakes signal that major rights-holders recognize the transformative potential of AI music, and want upside exposure, rather than merely seeking to suppress it.

To be clear, deals like the Warner Music Group-Suno partnership announced in November 2025 represent significant progress. Both parties have positioned it as a “blueprint for a next-generation licensed AI music platform” (Warner Music Group, 2025). The requirement for licensed models, artist opt-in controls, and deprecation of prior unlicensed systems addresses many immediate concerns. Yet even this landmark deal operates within existing copyright frameworks rather than resolving the deeper structural questions about how value should flow when influence becomes untraceable and outputs become infinite. As more such deals emerge, the industry will need to move beyond bilateral agreements toward sector-wide standards.

The Core Problem: Attribution No Longer Maps Cleanly

Traditional music licensing relies on identifiable lineage. A cover version derives from a specific song. A remix derives from particular stems. A sample derives from a master recording. These relationships are traceable, discrete, and legally definable.

Generative AI breaks this chain entirely.

When an AI model generates a song, it is not pulling from a single work, artist, or catalog. It synthesizes patterns learned across vast and diffuse datasets. The U.S. Copyright Office, in its May 2025 report on generative AI training, noted that AI models “require massive amounts of data” and that “a rudimentary model could be trained on a small music dataset,” but cutting-edge systems require training data measured in terabytes, potentially millions or billions of works (U.S. Copyright Office, 2025).

Suno itself has argued this point in court filings. In an August 2025 motion to dismiss a class action lawsuit brought by independent artists, Suno claimed that its AI “exclusively generates new sounds, rather than stitching together samples.” The company asserted that “no Suno output contains anything like a ‘sample’ from a recording in the training set, so no Suno output can infringe the rights in anything in the training set, as a matter of law” (Music Business Worldwide, 2025).

Whether or not courts accept this argument, it highlights the fundamental challenge: as AI systems improve, many outputs will no longer be meaningfully traceable to any specific source material. The question “who was this copied from?” increasingly becomes unanswerable, not because of bad faith, but because the premise of discrete derivation no longer applies.

The Fair Use Question: No Clear Consensus

The legal landscape remains deeply contested. In August 2024, both Suno and Udio acknowledged that they had trained their models on copyrighted recordings, but argued their practices constituted fair use. Suno claimed its model simply learned “the building blocks of music: what various genres and styles sound like.” Udio offered a similar defense, asserting that “musical styles, the characteristic sounds of opera, or jazz, or rap music, are not somehow proprietary” (Music Ally, 2024).

The RIAA rejected this reasoning categorically. “After months of evading and misleading, defendants have finally admitted their massive unlicensed copying of artists’ recordings,” the organization stated. “There’s nothing fair about stealing an artist’s life’s work, extracting its core value, and repackaging it to compete directly with the originals” (Music Ally, 2024).

The U.S. Copyright Office weighed in with its most definitive guidance in May 2025. The prepublication version of Part 3 of its AI report concluded that “making commercial use of vast troves of copyrighted works to produce expressive content that competes with them in existing markets, especially where this is accomplished through illegal access, goes beyond established fair use boundaries” (Crowell & Moring, 2025). The Office rejected arguments that AI training is inherently transformative, noting that AI models absorb “the essence of linguistic expression” in ways fundamentally different from human learning (Skadden, 2025).

However, the Copyright Office explicitly declined to recommend compulsory licensing or broad statutory changes. Instead, it favored allowing voluntary licensing markets to develop, noting that stakeholders generally prefer an “opt-in” approach where creators can choose when, how, and to whom they license their works (Jones Day, 2025).

Why Extending Existing Licensing Frameworks Will Ultimately Fail

Efforts to force AI-generated music into existing categories (covers, remixes, sampling) risk solving yesterday’s problem. These frameworks rest on three assumptions that generative AI fundamentally undermines:

First, they assume discrete source material. Traditional licensing identifies specific works being used. AI training involves statistical learning across enormous datasets where no single source dominates or can be isolated.

Second, they assume identifiable rights-holders per output. When a cover song is released, we know which composition was covered. When an AI generates a jazz track, there may be no meaningful answer to “which jazz recordings contributed to this output”; the output emerged from learned patterns across thousands of recordings.

Third, they assume direct derivation. Existing frameworks treat creative outputs as trees with traceable roots. AI outputs are better understood as emergent properties of complex systems; the whole is genuinely different from the sum of its parts.

We have already crossed this bridge in text. Large language models generate original prose informed by countless prior works, yet there is no practical mechanism to assign ownership to any individual author whose work contributed to the model’s training. Music is heading in the same direction, just with higher economic and cultural stakes.

Toward New Licensing Models

If the old rulebook no longer fits, what might replace it? Several models offer promising alternatives that better align with how generative AI actually functions.

1. Model-Level Licensing, Not Output-Level Licensing

Instead of attempting to license individual songs or outputs, rights-holders would license AI models themselves. Payments would be tied to model usage, training data participation, or commercial deployment, not attribution at the output level.

This is no longer theoretical. In November 2025, Warner Music Group and Suno announced what both companies called a “groundbreaking partnership” that demonstrates exactly how model-level licensing works in practice (Warner Music Group, 2025). The deal settled previous litigation between the companies and established a blueprint for licensed AI music platforms.

The agreement requires Suno to build a new generation of AI models trained exclusively on licensed music from Warner’s catalog. When these new models launch in 2026, Suno’s current models, which were trained on unlicensed data, will be deprecated entirely (Music Ally, 2025). This represents a fundamental shift: rather than fighting over whether past training constituted infringement, the parties agreed to build forward on a licensed foundation.

Warner Music CEO Robert Kyncl framed the deal in terms of principles: “AI becomes pro-artist when it adheres to our principles: committing to licensed models, reflecting the value of music on and off platform, and providing artists and songwriters with an opt-in for the use of their name, image, likeness, voice and compositions in new AI songs” (Rolling Stone, 2025).

Crucially, the partnership gives artists and songwriters full control over whether and how their names, images, likenesses, voices, and compositions are used in AI-generated music. Those who opt in gain access to new revenue streams; those who decline are protected. Suno CEO Mikey Shulman described the arrangement as enabling users to “build around participating artists’ sounds and ensure they get compensated” (Suno, 2025).

The deal also included Suno’s acquisition of Songkick, Warner’s concert-discovery platform, signaling an intent to connect AI-generated music with live performance ecosystems. With over 100 million users and a valuation of $2.45 billion following a $250 million funding round, Suno is the largest AI music platform in the world (Music Ally, 2025). This deal establishes the template that other major labels are likely to follow.

The Warner-Suno partnership validates what copyright advocates have long argued: legal pressure can compel AI companies to transition to licensed data models. Ed Newton-Rex, a vocal critic of exploitative AI training practices, has pointed to such deals as proof that litigation works (Marketing AI Institute, 2025). The question now is not whether licensing will happen, but how quickly the entire industry will follow this model.

Yet these deals are not without critics. The Music Artists Coalition, a nonprofit founded by legendary artist manager Irving Azoff, has cautiously welcomed the recent wave of AI agreements while raising pointed questions about their terms. “We’ve seen this before,” Azoff stated following the Universal-Udio settlement. “Everyone talks about ‘partnership,’ but artists end up on the sidelines with scraps” (Rolling Stone, 2025). Independent musicians and smaller labels not covered by major label agreements remain particularly vulnerable; their work may still have been used to train earlier models, and they lack the leverage to negotiate comparable protections. The opt-in framework, while preferable to no consent at all, may still favor established artists who command attention while leaving emerging creators with limited bargaining power.

This approach has historical precedent in how radio licensing evolved. In the early twentieth century, radio stations faced the impossible task of negotiating individual licenses with every composer and publisher whose music they might play. The solution was the blanket license: a single annual fee that grants access to an entire catalog.

ASCAP, founded in 1914, and BMI, established in 1939, pioneered this model (Wikipedia, 2025). Today, ASCAP’s blanket license provides access to over 20 million works from more than 1.1 million members (ASCAP, 2025). BMI tracks public performances from over 22.4 million musical works, collecting $1.573 billion in revenues and distributing $1.471 billion in royalties in fiscal year 2022 (Wikipedia, 2025). The blanket license “saves music users the paperwork, trouble and expense of finding and negotiating licenses with all of the copyright owners,” as ASCAP describes it (ASCAP, 2025).

AI training presents a parallel challenge requiring a parallel solution. Attribution at the output level is neither technically reliable nor economically scalable. Model-level licensing accepts this reality.

2. Training Rights as a Distinct Asset Class

Rather than treating AI training as an awkward extension of reproduction rights, training rights could be recognized as their own category. Artists and labels would opt in (or out) of training pools. Compensation would be based on dataset inclusion and model commercial success, not individual outputs.

In October 2025, Spotify announced a partnership with all three major labels plus independent label collective Merlin and distributor Believe to develop “artist-first AI music products.” The collaboration emphasized four principles: partnerships through upfront agreements, choice in participation (allowing artists to opt in or out), transparency about AI use, and fair compensation (Spotify Newsroom, 2025). This opt-in approach aligns with the Copyright Office’s observation that stakeholders generally prefer choosing “when, how, and to whom they license their works.”

This model reframes AI training as infrastructure investment rather than exploitation. Just as artists license synchronization rights for film and television (a right that did not exist until cinema created the need), training rights could become a recognized category with its own market dynamics.

3. Cultural Levy or Usage Pool

AI-generated music platforms could contribute a percentage of revenue into a collective pool, distributed to artists, labels, and publishers via industry bodies. Allocation could be weighted by historical consumption, influence metrics, or catalog participation.

This approach has substantial precedent in the private copying levies used across Europe. Under the EU’s 2001 InfoSoc Directive, twenty-two EU countries impose levies on blank media, eighteen on MP3 players, and twelve on printers (Kretschmer, 2011). In 2019, private copying fees accounted for 13% of the royalties European creators received from their collecting societies (GESAC, 2023). France collects approximately €2.60 per capita annually through this mechanism (Kretschmer, 2011).

The philosophical justification is similar: when technology makes individual tracking impractical, a levy system provides collective compensation. European collecting societies distribute these funds not only to individual rights-holders but also to cultural activities including concerts, festivals, and emerging artist support programs (GESAC, 2023).

A cultural levy on AI-generated music acknowledges that influence is diffuse without pretending precision is possible. It provides ongoing revenue streams to the creative ecosystem that AI systems draw upon.

4. Clear Separation Between AI-Native and Artist-Driven Works

Instead of obsessing over whether AI outputs are “derived” from human works, the industry could focus on intent and positioning. Artist-driven works, those led by human creativity and built around artist identity, would remain protected and marketed as human creations. AI-native works would be treated as a distinct category with different economics, expectations, and possibly different royalty structures.

Deezer has pioneered this approach through transparency. In June 2025, it became the first major streaming platform to explicitly tag AI-generated music. Its detection technology can identify fully synthetic content from major generators like Suno and Udio. The company removes AI-generated content from algorithmic recommendations and editorial playlists, and excludes fraudulent streams from royalty payments (Deezer Newsroom, 2025).

Consumer research supports this distinction. In Deezer’s November 2025 survey, 80% of respondents said AI-generated music should be clearly labeled for listeners. 45% said they would like AI content filtered out of their streaming service entirely. 69% agreed that payouts for fully AI-generated music should be lower than for human-made music (Deezer/Ipsos, 2025).

This approach avoids forcing AI music to compete directly with human artist identity while preserving space for human creativity to retain economic value.

The Economic Reality We Cannot Ignore

The music industry is already under considerable economic strain. Spotify, which commands approximately 31% of the global streaming market, pays artists an average of $0.003 to $0.005 per stream, meaning an artist needs roughly 200,000 to 333,000 streams to earn $1,000 (TuneCore, 2025). While streaming now accounts for 84% of U.S. music industry revenue (Royalty Exchange, 2025), the per-unit economics favor volume over individual compensation.

Starting in 2024, Spotify introduced a minimum threshold of 1,000 streams in the previous twelve months before any track generates royalties on the platform. According to Spotify, 99.5% of streams come from tracks meeting this threshold, but the policy effectively demonetizes vast numbers of smaller artists (iMusician, 2025).

AI will likely compress these economics further. More music means more abundance. More abundance means less unit value. Deezer found that approximately 70% of streams of AI-generated music on its platform were fraudulent: fake artists using bots to generate fake streams for payouts (NPR, 2025). Spotify removed over 75 million “spammy” AI tracks in 2024 alone (Music Business Worldwide, 2025).

This trend is uncomfortable, but it is not unprecedented. Similar compression occurred in print publishing, photography, graphic design, and video production as digital tools lowered barriers to entry. In each case, the economic center of gravity shifted, sometimes toward live experience, sometimes toward scarcity and authenticity, sometimes toward new business models entirely.

In this environment, live performance, community engagement, and documented human presence may become the primary domains where artists sustainably thrive. That is not a failure of policy; it is a structural rebalancing that policy must acknowledge rather than resist.

The Copyrightability Question

Beyond licensing, a parallel question looms: can AI-generated music receive copyright protection at all?

The U.S. Copyright Office addressed this in Part 2 of its AI report, published in January 2025. The Office affirmed that purely AI-generated material cannot be copyrighted: copyright protection requires sufficient human authorship. However, AI-assisted works can qualify for protection where “a human author has determined sufficient expressive elements” (U.S. Copyright Office, 2025).

The Office drew a crucial distinction: “using AI to assist in the process of creation or the inclusion of AI-generated material in a larger human-generated work does not bar copyrightability” (U.S. Copyright Office, 2025). The key question is whether AI enhances human expression or becomes the source of expressive choices.

This creates an interesting economic dynamic. Fully AI-generated works may flood platforms and dilute royalty pools, but they cannot receive copyright protection. Works with meaningful human involvement remain protectable. This bifurcation may ultimately reinforce the value of human creativity even as synthetic alternatives proliferate.

Conclusion: Rewrite the Rulebook, Don’t Patch It

Distinguishing AI music from human music, or extending existing licensing categories to cover AI outputs, will not be sufficient. The industry is attempting to apply rules designed for a world of scarcity to a future defined by abundance.

Protecting creators in the age of AI will require:

• Accepting that precise attribution has technical and conceptual limits
• Developing licensing systems that operate at model scale rather than output scale
• Creating new economic models that prioritize participation over provenance
• Establishing clear separation between human-led and AI-native creative categories
• Building collective compensation mechanisms for diffuse creative influence

The shift is going to happen whether the industry is ready or not. The major labels’ rapid pivot from litigation to licensing demonstrates they understand this. The question is whether the broader ecosystem, including independent artists, songwriters, performers, and publishers, adapts deliberately or defensively.

The music industry has survived technological disruption before: from player pianos to phonographs, from radio to MTV, from Napster to streaming. Each transition required not just legal adaptation but conceptual reimagining. The relationship between creativity, compensation, and technology had to be renegotiated from first principles.

We are in such a moment now. The old rulebook was written for a different technology, a different economy, and a different relationship between human creativity and machine capability. It served well for decades. It is time to write a new one.

References

• ASCAP. (2025). ASCAP Music Licensing FAQs. Retrieved from https://www.ascap.com/help/ascap-licensing
• Bloomberg. (2025, June 1). Universal, Warner, Sony in Talks to License AI Music Generators Suno and Udio.
• Crowell & Moring. (2025, May). U.S. Copyright Office Releases Third Report on AI and Copyright Addressing Training AI Models with Copyrighted Materials.
• Deezer/Ipsos. (2025, November 12). Survey on Perceptions and Attitudes Towards AI-generated Music.
• Deezer Newsroom. (2025, September 11). 28% of all delivered music is now fully AI-generated.
• GESAC. (2023). Private copying compensation. Retrieved from https://authorsocieties.eu/policy/private-copying/
• iMusician. (2025, February 14). How Much Does Spotify Pay Per Stream.
• Jones Day. (2025, May 22). U.S. Copyright Office Issues Guidance on Generative AI Training.
• Kretschmer, M. (2011). Private Copying and Fair Compensation: An Empirical Study of Copyright Levies in Europe. UK Intellectual Property Office.
• Level Law. (2025, October 28). Why the Suno Lawsuit Matters for the Music Tech Ecosystem.
• Marketing AI Institute. (2025, December 4). Music Battle Ends, New Partnership Begins with Suno and Warner Music.
• Music Ally. (2024, August 2). Suno and Udio slam label lawsuits… but the RIAA hits back.
• Music Ally. (2025, November 25). AI-music firm Suno strikes first licensing deal… with Warner Music Group.
• Music Business Worldwide. (2025, August 26). Suno argues none of the millions of tracks made on its platform ‘contain anything like a sample.’
• Music Business Worldwide. (2025, November 12). 50,000 AI tracks flood Deezer daily.
• NPR. (2025, August 8). AI-generated music is here to stay. Will streaming services like Spotify label it?
• RIAA. (2024, June 24). Record Companies Bring Landmark Cases for Responsible AI Against Suno and Udio.
• Rolling Stone. (2025, November 26). AI-Music Heavyweight Suno Partners With Warner Music Group After Lawsuit Settlement.
• Royalty Exchange. (2025). How Music Streaming Platforms Calculate Payouts Per Stream 2025.
• Skadden. (2025, May). Copyright Office Weighs In on AI Training and Fair Use.
• Suno. (2025, November 25). A new chapter in music creation. Retrieved from https://suno.com/blog/wmg-partnership
• Spotify Newsroom. (2025, October 16). Sony Music Group, Universal Music Group, Warner Music Group, Merlin, and Believe to Partner With Spotify to Develop Artist-First AI Music Products.
• TuneCore. (2025). How Much Does Spotify Pay Per Stream in 2025.
• U.S. Copyright Office. (2025, January 29). Copyright and Artificial Intelligence, Part 2: Copyrightability.
• U.S. Copyright Office. (2025, May 9). Copyright and Artificial Intelligence, Part 3: Generative AI Training. Pre-publication version.
• Variety. (2025, November 20). Universal, Warner and Sony Strike Licensing Deals With AI Music Startup Klay.
• Warner Music Group. (2025, November 25). Warner Music Group and Suno Forge Groundbreaking Partnership. Press Release.
• Wikipedia. (2025). Broadcast Music, Inc.