An artist can agree to an AI license without being able to verify a single payment under it.
An AI license is the permission a rights holder gives an AI company to use recordings or compositions, most often as training material for a model that generates new music, and sometimes for what the model produces or for imitation of a performer's voice.
In arrangements such as Believe's partnership with Suno, a distributor sits between participating artists and the AI company. The announcement promises choice and compensation, but does not explain what artists will see before consenting or how they will verify their payments afterward. Without access to the relevant terms and supporting records, a payment can become a number on a statement with no visible path back to the agreement that produced it. Obviously, that's a problem and that problem has three parts to it: consent, compensation and verification.
Consent addresses whether the use is authorized. Compensation determines what the permission is worth and how the money is calculated. Verification means the reporting and audit rights that let the person owed money check whether the agreement was honored. Recent AI deal announcements promise consent and compensation but explain little about verification.
The third part makes the other two enforceable in practice. A right to payment is harder to exercise without the ability to check the calculation. Our earlier article on the So So Def complaint examined the consequences of reconstructing royalty accounting decades later: disputed statements, amended accounts, and catalog values that turn on incomplete records. AI licensing is a chance to build that record while the revenue stream is still being designed.
The goal is a contract connecting the permission granted, the income received, and the calculation of the artist's share. Without that connection, even identifying an underpayment can require a costly reconstruction.
Distributors can earn on both ends
On September 8, 2026, Believe announced a partnership with Suno. Believe provides distribution and label services to independent artists and labels; TuneCore is its self-service distribution platform. Suno develops generative AI software that produces songs from prompts. The partnership covers participating Believe and TuneCore repertoire, with compensation and rights protection promised to artists who consent. Tracks created with Suno's new industry-partner model will also become eligible for distribution through Believe and TuneCore.
This is not the first independent-side AI agreement. Merlin, which negotiates digital licenses for independent labels and distributors, announced its Udio partnership in January 2026. What makes Believe's arrangement instructive is its reach into self-releasing artists and its two-way structure: repertoire flows into the model, and generated music can flow back through the distributor.
Those are different businesses. Compensation for supplying recordings is one revenue stream; fees for distributing generated tracks are another. An artist weighing the opt-in should understand which activity generates the distributor's return, or whether it earns from both, because those interests shape the opportunity being presented. TuneCore's program terms, discussed below, show the distributor retaining a share in at least two categories of AI receipts.
Believe's announcement does not explain participant-level calculations or inspection rights. Neither does BMG's August 12 Suno announcement. That silence does not establish that the private agreements lack protections. It establishes what an artist cannot learn from the announcements alone.
Know what you are licensing and who can approve it
Following the money starts with knowing what was licensed. The phrase AI licensing can cover at least four permissions.
Training uses recordings and compositions as material a model learns from, including copies assembled in datasets. Output use concerns whether protected material may appear in generated music. Voice and likeness concerns imitation of a performer, raising legal questions distinct from copyright in the recording or song. Distribution concerns how generated tracks reach listeners. These permissions can carry different terms, durations, and prices. Consent to one should not be assumed to include the others.
Time is another boundary. BMG expressly says its agreement settles prior use of recordings and compositions. Believe's announcement does not address a prior-use settlement. Permission for a new model should not be read as resolving earlier claims unless the contract says so.
The person granting permission must also have the necessary authority. A company may own rights, administer them for an owner, or hold a limited license for distribution. A license to deliver recordings to streaming services does not, by itself, establish authority to supply a training dataset. The artist's agreement determines that scope.
Copyright adds a separate distinction. A sound recording and the composition it embodies are separate works, often with different owners. The recording, or master, may be controlled by a label or artist; the underlying song by songwriters and publishers. Authority over one does not establish authority over the other.
Collaborators complicate the picture further. A producer's points, meaning a share of recording income, do not necessarily confer approval rights. A featured performer's approval does not cover a songwriter's interest. A collecting society, meaning an organization that licenses and collects on behalf of rights holders such as songwriters and publishers, may also hold a mandate that needs to be accounted for. Contracts and applicable law determine whose consent is needed and whose financial participation follows. Those are different questions, and both belong in the licensing record.
TuneCore uses three different payment rules
Once permission and authority are clear, the oldest question in royalty accounting applies: a percentage of what?
A rate cannot explain a payment without its base. Net receipts means income after the deductions the contract permits. The definition of income, the deductions, and the category assigned to a receipt can matter as much as the headline percentage.
TuneCore's AI and Data Protection Program terms, dated November 26, 2024, make this concrete. The framework for offering participating content to AI partners contains three compensation rules.
Recoveries: artists share what remains after deductions and TuneCore's cut. If TuneCore recovers money for unauthorized dataset use, an artist does not simply receive half of the settlement. Section 1.4 authorizes these Dataset Claims; section 2.1 allows defined deductions, including reasonable expenses and legal fees, before TuneCore retains half. The remainder is divided among participants according to content use or other appropriate criteria within TuneCore's good-faith discretion. Both the deductions and the allocation determine the artist's payment.
Limited launches: if TuneCore receives nothing, the artist agrees not to claim payment. An Experimental Launch is a limited release rather than a launch at scale. Section 2.2 provides a 50% share of defined net receipts. But if TuneCore "does not receive any sum," the artist waives "all rights to claim any fees and/or royalties from TuneCore and any Partner" for that launch. Participation can therefore mean authorizing an experimental use without compensation.
Full launches: the rate is set later, in terms presented for approval. For a Formal Launch, meaning a launch at scale, section 2.3 leaves the economics to dedicated terms presented for participant approval. The experimental-launch percentage does not carry over as a promised commercial rate.
These are program terms. The announcements do not disclose Suno-specific launch terms, and those terms were not reviewed for this article. Dedicated launch terms may add or change provisions. Even so, the program shows why the type of use and the category of receipt must be understood before a percentage can tell the artist anything useful.
How the money is divided determines your share
When a payment identifies a particular recording and use, it provides a starting point for matching income to the applicable contract. A lump sum covering a catalog's training rights or settling prior use may not arrive with that breakdown. Someone must divide it among participants. Unless the contract fixes the method, the artist may be accepting discretion over a central part of the calculation.
Allocation is not inherently improper. The UK Competition and Markets Authority's 2022 streaming study, in footnote 263, described non-artist-specific receipts typically being passed to artists through allocation rules, including an artist's share of a label's streams on a service. The question is whether the rule fits the agreement and can be checked.
Suppose an AI payment is divided by historical streaming share. That rewards commercial popularity, not necessarily contribution to a model. An attribution system, by contrast, might estimate which source works influenced an output. Its inputs and limitations need scrutiny: an estimate depends on the data and assumptions behind it. Neither approach becomes the correct contractual allocation merely because software produces a number.
The artist needs to know not only the share of the pool, but how the pool was defined and how that share was assigned.
A statement is not an audit right
A royalty statement is the payer's summary of its calculation. An audit right allows the payee to examine the underlying books and records, usually through an independent accountant. An objection period sets the time available to challenge a statement. Together, these provisions determine whether an artist can test the promise of compensation.
TuneCore's AI terms expressly incorporate its Music Distribution terms, dated July 31, 2026, into the program's Terms of Service. Those provisions state that the artist "shall have no right to inspect or audit Company's books and records or the books and records of any Consumer Store," and they make statements "conclusively binding" one year after they are rendered. The restriction is incorporated into the program, not merely located in an unrelated document.
A participant therefore needs to check whether dedicated launch terms override those restrictions or add access. Three questions make that review concrete:
- Can the artist understand the deal before consenting? The operative permissions, payment provisions, deductions, allocation rules, and restrictions should be available, along with supporting license terms where needed to understand or verify them.
- Does reporting travel with the money? The AI company should supply the information its licensor needs to calculate participants' shares, and the licensor should pass enough detail downstream to explain each participant's statement.
- Can an audit reach the calculation? Inspection should cover relevant upstream reports, allocation inputs, and deductions, rather than stop at the arithmetic on the final statement.
This does not require publishing every commercial term. Confidential access for a qualified independent auditor can protect sensitive information while allowing verification. Record retention and realistic objection periods matter because a right to inspect records is of little use after they disappear.
Identification tools solve a different problem. Suno's August 2026 responsible-AI announcement describes planned watermarking and fingerprinting to help identify Suno-generated music. Those tools address provenance: where a track came from. Accounting asks what a particular rights holder is owed. Establishing the first does not establish the second.
Disclosure should not begin with a lawsuit
GEMA's case against Suno shows disclosure operating as a remedy. GEMA represents composers, lyricists, and music publishers. The Munich I Regional Court's July 31, 2026 judgment ordered Suno to disclose specified uses and associated revenues concerning particular works. The dispute concerned versions 3.5 and 4, and the published record identifies an appeal. It is not a final, worldwide rule about AI training or a ruling on the new licensing contracts.
Its relevance here is narrower: records of use and revenue formed part of the remedy. A licensing agreement can require relevant reporting routinely, rather than leave a rights holder to seek disclosure through litigation.
In transactional work, I review a revenue line by asking what document creates it, what governs its division, and what records would prove the division was correct. AI income deserves that review before the first payment, not after the first disagreement.
That is the approach behind Cadence. Reading the contract establishes the permissions, receipt categories, deductions, and allocation rules. Connecting those terms to source reports and payment histories creates a record that can explain how a statement was produced. For an independent label or distributor, this supports both checking receipts from an AI partner and accounting to its own artists. For an artist, it is the difference between receiving a number and understanding it.
AI licensing adds new categories to this work. It does not remove the need for it. Artists should not have to choose between participating in a new market and understanding their place in its economics.