AI Music Deals: Who Gets to Say Yes — or No?

The argument over artificial intelligence and music is often presented as a fight between creativity and technology. The more practical question is narrower: what exactly is being licensed, who has authority to license it, and did the person who made it get a meaningful choice?

That question became harder to answer as record labels moved from suing AI music companies to making deals with some of them. The Los Angeles Times reported in September that Suno had launched models built with licensed music from Warner Music Group, BMG and Believe, while Udio had deals with Universal Music Group and Warner. The companies describe licensing as a way to build lawful products and return some revenue to music rights holders. But important terms, including what artists are paid and how many have opted in, have not been made public.

For listeners, the debate can sound abstract. For a singer, songwriter, or musician hired for a recording session, it can affect whether one performance remains a paid piece of work—or becomes raw material for thousands of new sounds.

Four different things people mean by “the rights”

A master recording is a particular recorded performance: the audio file you hear on a finished track. A record label commonly owns or controls this copyright, depending on the artist’s contract. Licensing a catalog to an AI platform may therefore give a label permission to use those recordings in training, product features, or both. Those are separate questions, and an opt-out may not always mean the same thing in each setting.

A composition is the underlying song: its melody, lyrics and musical structure. Songwriters and music publishers commonly control this side. A label’s permission to use a sound recording does not automatically settle every question about the composition. The Times reported that songwriter advocates see this side as especially difficult to police because a copied recording can be recognizable, while influence on a song’s writing may be more subtle.

Voice and likeness concern identity. A generated track that imitates a recognizable singer may raise questions even when it does not copy a particular master recording. Billboard reported that Sony had requested removal of more than 260,000 AI deepfake tracks that it said imitated its artists’ voices and likenesses without permission. The United States still lacked a single federal law covering every use of a person’s name, image, likeness or voice in an AI deepfake, leaving an uneven mix of state protections at the time of that report.

Labor contracts govern the work people perform and the terms attached to it. A session musician may have played a horn, guitar, drum or backing vocal on a recording without owning the master. That does not necessarily answer whether a later AI use of that performance was included in the original bargain. Contracts, union agreements and payment rules can determine what a player was promised, but the wording may predate generative AI by decades.

Keeping these categories separate matters. A label may have authority over a master while a songwriter, performer or session player has a different interest that still needs to be addressed. Calling everything “the music” can hide who is being paid—and who is being asked to accept a new use.

Why labels and AI companies see a workable path

The industry case for licensing is not simply that technology is inevitable. Labels and AI companies say negotiated deals can replace unauthorized scraping and imitation with defined rules. Licensed catalogs can help platforms build tools with permission, attribution and payment systems instead of leaving artists to chase copies after they appear online.

The deals also promise a new revenue stream. The Times reported that participating companies describe products in different ways: Udio’s planned “walled garden” limits how generated songs can be distributed, while Suno allows downloads subject to limits and watermarks. A platform may also create opportunities for independent artists through distribution partners. In that view, licensing is a way for rights holders to participate in a market that is developing whether they welcome it or not.

There is a reasonable business argument here. Rights owners already negotiate permissions for streaming, synchronization and other uses. A properly scoped AI license could specify what material is used, how it is identified, how revenue is calculated, and what happens when a creator says no. It could also give platforms a clearer legal foundation and give listeners tools that are less dependent on unauthorized celebrity soundalikes.

But “a label made a deal” and “each artist made a choice” are not identical statements. The Times reported that major-label contracts may give companies substantial control over an artist’s music, while the exact effect of an opt-out can remain unclear. A revenue stream can be real and still be distributed unevenly.

Why artists and session players remain uneasy

Artist and songwriter groups say consent should be direct, informed and reversible—not buried in a new clause or inferred from an old agreement. In June, a coalition including the Music Artists Coalition, Songwriters of North America, NITO and BMAC objected to default AI opt-ins and asked for fair compensation, transparency and an end to forced clauses. Their concern is not necessarily opposition to every AI tool. It is that a person may lose the ability to refuse without losing a recording or publishing deal.

Session musicians face a particularly practical worry. The Times reported that the American Federation of Musicians sued Universal and Warner, alleging that recordings featuring its members were licensed to Suno and Udio without the musicians being paid or told which recordings were involved. The labels moved to dismiss the case. Those allegations remain contested, but they illustrate the gap between owning a master and paying the people whose labor helped create it.

For a working player, replacement is not only a philosophical fear. If a single old session can be used to generate convincing instrument parts, future employers may need fewer live players—or may pay less for their time. The same technology could also help musicians sketch ideas, make demos or reach audiences. The outcome depends on the contract, the product and whether human workers share in the value.

Transparency is therefore more than a public-relations preference. Artists need to know which recordings or compositions are covered, whether training and user-facing generation are treated differently, how money is counted, and whether a generated track can imitate a person’s identity. Without that information, “choice” may exist on paper while remaining difficult to exercise.

What a meaningful right to say no would require

A credible opt-out would need to answer four simple questions: No to what use? For how long? With what payment? And with what remedy if the material reappears? It should distinguish a label’s rights from a performer’s voice, a songwriter’s composition and a session player’s labor. It should make the terms understandable before a person signs, not after a dispute begins.

There is no single outcome that fits every artist. Some may welcome licensed AI collaborations, new income or creative tools. Others may reject synthetic versions of their voices or worry that agreeing today weakens their livelihood tomorrow. Both positions can be legitimate, particularly when bargaining power is unequal.

The public test is not whether a deal carries an “AI” label. It is whether the people whose performances, writing and identities make the deal valuable can see the terms, share in the money, and decline a use without being quietly treated as obstacles. Until payouts, opt-in numbers and contract details become clearer, it is too soon to call the new licensing model either a rescue for musicians or a certain route to their replacement.

Sources and further reading

• Los Angeles Times: “Why some musicians aren’t happy about labels signing AI deals with platforms” (Sept. 21, 2026)

• Los Angeles Times: “‘The ability to say no’: Artists demand consent over AI use of their music” (June 23, 2026)

• Billboard: “Sony Music Has Requested the Removal of Over 260,000 AI Deepfake Songs From Digital Platforms” (Oct. 5, 2026)

• American Federation of Musicians press room

Update note: This explainer reflects reporting and public statements available on October 7, 2026. The lawsuits, licensing terms and proposed legal protections may change.