Article 21(a) of the SRLA says a company “shall pay” musicians when a covered recording is used for a purpose outside the agreement. Universal and Warner argue the clause still needs another AFM agreement to supply the rate, which is where the AI dispute gets unusually technical.
No AFM agreement was written specifically for generative AI training when the current Sound Recording Labor Agreement took effect in 2023. The labels treat the gap as decisive because, in their reading, Article 21 is a conversion mechanism that borrows a rate from an agreement already governing the later use.
The union reads the same sentence differently. Its position in the AFM contract fight over AI licensing is that the duty to pay can arise first, while the amount can be worked out from existing AFM rates or determined later.
Universal’s position effectively ties those two parts together. If no separate agreement already covers AI training, there is no rate to import, so the payment provision has nothing to calculate and the claim fails. Warner reaches the same practical destination by arguing Article 21 points to other agreements rather than creating a freestanding payment right.
The AFM is trying to pull those steps apart. In its reading, “shall pay” creates the obligation once the new use occurs, while the later language tells the parties how to measure what is owed. A missing bespoke AI schedule could therefore create a damages problem without automatically erasing the underlying payment duty.
Contract scholarship recognizes a broader version of this problem. Research on ambiguous contracts examines agreements capable of supporting more than one payoff structure, although a federal labor case turns on legal interpretation rather than an economics model. Here, the practical issue is whether one clause can carry a binding trigger even when its pricing mechanism is disputed.
This wording gives the union another route around the missing-rate argument. A recording prepared for use by an AI platform could still be a covered phonograph record, so the AFM says the SRLA itself may remain the relevant agreement for the musician’s recorded work even if no separate AI platform agreement exists.
Article 21 also contains more plumbing than the phrase “new-use clause” suggests. Paragraph 21(a) works as the general rule, while later paragraphs carve out particular situations and can use percentage-of-license-fee formulas instead of the full payment method. The complaint describes some of those later provisions as tying particular uses to licensing revenue, so Article 21 already accommodates more than one payment route.
Video game uses are among the examples the AFM has pointed to when arguing that new technology did not always arrive with a neat, prewritten rate card. The union says companies treated early game licensing as a new use before a dedicated AFM agreement had supplied a rate for it.
A judge could still accept the labels’ reading. Yet “there is no AI rate” is less tidy than it sounds because the dispute also turns on which agreement counts, what kind of recording is being used, and whether a pricing gap affects liability or only the eventual amount.
Downstream uses matter too. AI-generated recordings can be streamed, sampled, placed in games, or used commercially, and some of those activities sit closer to negotiated AFM rate structures than model training does. One AI license can therefore touch several economically different uses rather than one clean category called “AI.”
Nobody needs to pretend the 2023 agreement contains a bespoke generative AI training price. It does not. The harder issue is whether Article 21 was built to stop compensation from disappearing whenever a genuinely new use arrives before bargaining catches up, leaving the amount for evidence, interpretation, and later calculation.
No AFM agreement was written specifically for generative AI training when the current Sound Recording Labor Agreement took effect in 2023. The labels treat the gap as decisive because, in their reading, Article 21 is a conversion mechanism that borrows a rate from an agreement already governing the later use.
The union reads the same sentence differently. Its position in the AFM contract fight over AI licensing is that the duty to pay can arise first, while the amount can be worked out from existing AFM rates or determined later.
Article 21 separates the payment trigger from the rate
Read the clause closely, and the split is easy to spot. Article 21(a) first addresses a covered phonograph record used for a purpose outside the SRLA, then describes the amount by reference to payments required under the AFM agreement that would be effective for the new purpose.Universal’s position effectively ties those two parts together. If no separate agreement already covers AI training, there is no rate to import, so the payment provision has nothing to calculate and the claim fails. Warner reaches the same practical destination by arguing Article 21 points to other agreements rather than creating a freestanding payment right.
The AFM is trying to pull those steps apart. In its reading, “shall pay” creates the obligation once the new use occurs, while the later language tells the parties how to measure what is owed. A missing bespoke AI schedule could therefore create a damages problem without automatically erasing the underlying payment duty.
Contract scholarship recognizes a broader version of this problem. Research on ambiguous contracts examines agreements capable of supporting more than one payoff structure, although a federal labor case turns on legal interpretation rather than an economics model. Here, the practical issue is whether one clause can carry a binding trigger even when its pricing mechanism is disputed.
The SRLA already uses technology-neutral recording language
The contract is not completely silent about technologies that did not exist when older versions of the agreement were negotiated. The SRLA definition cited by the AFM describes a phonograph record broadly enough to include a digital audio file and other devices reproducing sound, including forms that may come into existence later.This wording gives the union another route around the missing-rate argument. A recording prepared for use by an AI platform could still be a covered phonograph record, so the AFM says the SRLA itself may remain the relevant agreement for the musician’s recorded work even if no separate AI platform agreement exists.
Article 21 also contains more plumbing than the phrase “new-use clause” suggests. Paragraph 21(a) works as the general rule, while later paragraphs carve out particular situations and can use percentage-of-license-fee formulas instead of the full payment method. The complaint describes some of those later provisions as tying particular uses to licensing revenue, so Article 21 already accommodates more than one payment route.
Video game uses are among the examples the AFM has pointed to when arguing that new technology did not always arrive with a neat, prewritten rate card. The union says companies treated early game licensing as a new use before a dedicated AFM agreement had supplied a rate for it.
A judge could still accept the labels’ reading. Yet “there is no AI rate” is less tidy than it sounds because the dispute also turns on which agreement counts, what kind of recording is being used, and whether a pricing gap affects liability or only the eventual amount.
Existing rates could become evidence instead of a perfect match
The AFM has identified session, streaming, and sampling rates as objective measures that could inform compensation. Those categories are not identical to model training, but the union does not need them to be identical if the court accepts that Article 21 already creates the payment obligation.Downstream uses matter too. AI-generated recordings can be streamed, sampled, placed in games, or used commercially, and some of those activities sit closer to negotiated AFM rate structures than model training does. One AI license can therefore touch several economically different uses rather than one clean category called “AI.”
Nobody needs to pretend the 2023 agreement contains a bespoke generative AI training price. It does not. The harder issue is whether Article 21 was built to stop compensation from disappearing whenever a genuinely new use arrives before bargaining catches up, leaving the amount for evidence, interpretation, and later calculation.
