Suno signed licensing agreements with Warner Music Group, BMG, and Believe before Universal and Sony filed their second copyright case in September 2026. The labels now point to those deals as evidence that a commercial market exists for authorized use of recordings in generative AI.
The timing gives the argument some bite. Warner settled its Suno litigation and announced a forward-looking partnership in November 2025, BMG followed in August 2026, and Believe announced its agreement on September 8, one day before Suno launched v6.
Universal and Sony use those transactions inside Suno's second copyright case over 60,202 recordings to attack part of Suno's fair-use position. Their point is straightforward, but the legal question underneath it is much messier than counting how many companies signed deals.
BMG's August agreement covers its recorded-music and publishing repertoire within a broader strategic framework. BMG said participating artists and songwriters would be compensated and protected, while the arrangement also settled prior use of BMG recordings and publishing works.
Believe's deal is different again. It covers participating Believe and TuneCore repertoire, lets artists and labels opt in to licensed products, and makes tracks created with Suno's new industry-partner model eligible for distribution through Believe and TuneCore.
Those differences matter because the public announcements do not reveal a clean market price for training one recording. Settlement value, access to repertoire, future products, artist participation, distribution, publicity rights, and other commercial rights can sit inside the same agreement, which makes a headline licensing deal a poor substitute for an itemized rate card.
There is an important limit. Courts have warned that fair use cannot be defeated simply by defining the relevant market as whatever fee a copyright owner would have charged for the disputed use, because that reasoning can become circular.
A 2025 federal ruling in the Meta book-training litigation made the point sharply. The court accepted that a licensing market for AI training might exist or develop, yet rejected lost training-license fees as the decisive market harm on the record before it because the claimed market depended on a right to control the transformative use itself.
Massachusetts is not bound by that California district-court ruling, and Suno involves sound recordings rather than books. Still, economic analysis of generative AI copyright also treats compensation rules, creator incentives, AI development, and market effects as interacting questions rather than one switch that flips whenever a license exists.
Even so, later deals do not automatically establish what a lawful license would have cost when earlier Suno models were trained. Public terms do not disclose the economics needed to separate payment for past claims from payment for future training, product access, distribution rights, or other concessions.
Warner's agreement is especially awkward as a benchmark because it also ended litigation. A settlement can reflect litigation risk, bargaining leverage, confidentiality, and the value of peace, so its total economics may say more than the price of permission to ingest recordings.
BMG adds another wrinkle because its announcement says the deal settled prior use while creating future opportunities. Believe adds still another because its arrangement links participating repertoire with new licensed products and distribution eligibility, showing a commercial exchange broader than training access alone.
For Universal and Sony, the deals help demonstrate that rights holders and an AI music company can actually build paid, permission-based arrangements rather than merely theorize about them. For Suno, the unresolved question is whether those packages prove legally cognizable market harm from the earlier copying, and whether their mixed settlement and partnership terms can reliably measure the value of the specific rights now in dispute.
The timing gives the argument some bite. Warner settled its Suno litigation and announced a forward-looking partnership in November 2025, BMG followed in August 2026, and Believe announced its agreement on September 8, one day before Suno launched v6.
Universal and Sony use those transactions inside Suno's second copyright case over 60,202 recordings to attack part of Suno's fair-use position. Their point is straightforward, but the legal question underneath it is much messier than counting how many companies signed deals.
The three deals are not copies of the same license
Warner's partnership did two jobs at once. It settled Warner's earlier lawsuit and created a framework for new licensed Suno models, with Warner publicly stressing compensation, artist protections, and opt-in treatment for certain uses of artists' names, images, likenesses, voices, and compositions.BMG's August agreement covers its recorded-music and publishing repertoire within a broader strategic framework. BMG said participating artists and songwriters would be compensated and protected, while the arrangement also settled prior use of BMG recordings and publishing works.
Believe's deal is different again. It covers participating Believe and TuneCore repertoire, lets artists and labels opt in to licensed products, and makes tracks created with Suno's new industry-partner model eligible for distribution through Believe and TuneCore.
Those differences matter because the public announcements do not reveal a clean market price for training one recording. Settlement value, access to repertoire, future products, artist participation, distribution, publicity rights, and other commercial rights can sit inside the same agreement, which makes a headline licensing deal a poor substitute for an itemized rate card.
A licensing market does not automatically defeat fair use
Fair use weighs several factors, including the effect of the challenged use on the potential market for or value of the copyrighted work. Universal and Sony argue Suno's own transactions undermine any suggestion that licensing music for AI development is merely hypothetical.There is an important limit. Courts have warned that fair use cannot be defeated simply by defining the relevant market as whatever fee a copyright owner would have charged for the disputed use, because that reasoning can become circular.
A 2025 federal ruling in the Meta book-training litigation made the point sharply. The court accepted that a licensing market for AI training might exist or develop, yet rejected lost training-license fees as the decisive market harm on the record before it because the claimed market depended on a right to control the transformative use itself.
Massachusetts is not bound by that California district-court ruling, and Suno involves sound recordings rather than books. Still, economic analysis of generative AI copyright also treats compensation rules, creator incentives, AI development, and market effects as interacting questions rather than one switch that flips whenever a license exists.
The structure of the deals may matter more than the count
The labels have a stronger factual point than saying three contracts exist. Suno moved from defending unlicensed training as fair use to launching v6 with Warner, BMG, and Believe as industry partners, while those partners publicly described compensation and controlled participation as features of the new arrangements.Even so, later deals do not automatically establish what a lawful license would have cost when earlier Suno models were trained. Public terms do not disclose the economics needed to separate payment for past claims from payment for future training, product access, distribution rights, or other concessions.
Warner's agreement is especially awkward as a benchmark because it also ended litigation. A settlement can reflect litigation risk, bargaining leverage, confidentiality, and the value of peace, so its total economics may say more than the price of permission to ingest recordings.
BMG adds another wrinkle because its announcement says the deal settled prior use while creating future opportunities. Believe adds still another because its arrangement links participating repertoire with new licensed products and distribution eligibility, showing a commercial exchange broader than training access alone.
For Universal and Sony, the deals help demonstrate that rights holders and an AI music company can actually build paid, permission-based arrangements rather than merely theorize about them. For Suno, the unresolved question is whether those packages prove legally cognizable market harm from the earlier copying, and whether their mixed settlement and partnership terms can reliably measure the value of the specific rights now in dispute.