AI Music on Trial: Is Suno’s “Back-End” Use of Copyrighted Recordings Fair Use?

Is Suno’s “Back-End” use of copyrighted recordings fair use?

Answer: Suno’s “back end” use of UMG’s recordings is likely not fair use because its threat of market supplantation outweighs its transformative purpose.

On June 24, 2024, Universal Music Group and other affected labels (hereinafter, “UMG”) sued Suno, Inc. in the U.S. District Court for the District of Massachusetts. UMG alleges Suno committed copyright infringement by using proprietary sound recordings to train its artificial intelligence (AI) music model. UMG alleges that Suno infringed their catalog of copyrighted recordings because Suno used Plaintiff’s recordings to train Suno’s AI model. UMG offers AI outputs that replicate protected expressive features as evidence that copyrighted recordings exist in the back end of suno’s catalogue. In its answer, Suno contends that back-end copying for technical training constitutes fair use. Suno further asserts that sound recording rights protect only the recording itself, and that UMG only proved Suno replicated the underlying composition.

This memorandum objectively compares the allegations in UMG’ complaint with the defenses in Suno’s answer, focusing on whether AI "ingestion" constitutes transformative fair use or direct infringement

What is fair use?

Authors of original works own the exclusive right to exploit those works. Copyright infringement occurs when a party exploits a work without ownership or authorization. Fair Use is an affirmative defense that bars liability for otherwise infringing conduct to advance the creative arts and sciences.

Fair use is an equitable rule of reason that aims to promote the progress of science and useful arts. In finding fair use, the court uses a statutory four factor balancing test.17 U.S.C. § 107.The fair use balancing test weighs the following to determine if the derivative use promotes the progress of science and useful arts:

1. the purpose and character of the use,

2. the nature of the copyrighted work;

3. the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and

4. the effect of the use upon the potential market for or value of the copyrighted work

   
   

Fair Use decisions are a fact intensive, case specific inquiry. In each fair use decision, the judge weighs each factor against the other, in light of the case’s specific facts to make a determinative outcome. Thus, prior rulings on similar subjects provide predictive guidance on how a judge may weigh each factor. In recent AI litigation, judges weighed the transformative purposes of AI against the threat to the market value of the original work.

Suno faces significant hurdles in establishing a fair use defense, primarily due to the expressive nature of the source material and the risk of market displacement.

Courts will weigh “back-end” use’s transformative purpose against its potential for market competition.

While AI music presents a novel issue, recent district court decisions regarding AI trained on written materials provide persuasive guidance for how the higher courts may rule. In these cases, Courts have largely favored AI defendants on the first factor, concluding that training generative AI on copyrighted materials is "exceedingly transformative". However, some courts view the threat of market competition as the “most important factor in the fair use analyisis”, and balk at the unique threat that Suno poses.

In Bartz, the court analogized AI ingestion to a "voracious reader" ingesting classic books as they aspire to become a writer—a type of market competition the court declines to prevent. In Kadrey, the court argued that AI poses a significantly greater market threat than a “voracious reader” because AI can generate thousands of works rapidly without creative effort. Furthermore, the court noted that a transformative purpose is not a "silver bullet" for fair use. The Kadrey court argued that AI poses a significantly greater market threat than the reader in Bartz because AI can generate thousands of works rapidly without creative effort.

In sum, Courts will likely be sympathetic to the transformative nature of Suno’s use, but a compelling argument about market competition may prove fatal to Suno, depending on which thread the superior court follows.

Factor One: Purpose and Character of the Use

Suno’s derivative use has a different purpose or character than UMG because it has a transformative purpose. The first factor of a fair use inquiry examines the purpose and character of the derivative work. A work is more likely to be fair use when it serves a different purpose than the original rather than merely superseding it. A work that has a transformative purpose weighs heavily in favor of a fair use finding.

Suno’s derivative use of Plaintiff’s properties had a transformative purpose because it uses the original work to train back end technology. The use of proprietary materials to train back-end technology is transformative when the system does not display the source work to the public. Here, Suno’s purpose is transformative because their system never reveals the original copyrighted content to its users. In Authors Guild, the court found that the use of copyrighted books to provide data and content for the back end of a commercial search function was transformative because it created a user interface that utilized, but did not reveal, the entire work. Suno, by comparison, is even more transformative because their product never reveals untouched portions of the original work while the use in Authors Guild revealed entire untouched lines of the book.

The rulings in both Bartz and Kadrey show that courts may be persuaded that the distance between the source material and the end-user makes AI training transformative. Suno’s attorneys may further argue that musical works contain diverse creative elements—many of which UMG do not own—making the AI’s use even more transformative than the text-based cases.

UMG will argue Suno’s use is not transformative because AI song creation threatens to supplant the original works. A use that adds a new expression is not transformative if it shares the same commercial purpose as the original or otherwise threatens to supplant the market for the original. In Warhol, Plaintiff’s use of Prince’s portrait was not transformative because its purpose was similarly commercial. Here, Suno generates music at a scale that risks drowning out genuine recordings. Because Suno-generated tracks have already appeared on streaming services, they compete directly for royalties with the original records. Consequently, Suno's derivative use occupies the same purpose and character as the original works.

This factor hinges on where the court decides to analyze arguments about market substitution. While Warhol provides precedent for weighing such arguments against the purpose of the work, the lower courts in both Bartz and Kadrey address these concerns in their analysis of the fourth factor instead of embedding that conversation in its first factor analysis. It’s likely that the higher courts will follow their lead. The first factor favors a finding of fair use because the court will likely ignore arguments about market competition until later in their analysis.

Factor Two: The Nature of the Use

The second factor weighs against a finding of fair use because Plaintiff’s Copyrighted recordings are inherently expressive and subject to greater protections. Expressive works are afforded more protection than merely functional works. The main function of the second factor is to inform how the Court interprets the other factors. If the Court finds that the nature of the original work is expressive, a derivative work is less likely to be Fair Use because the Court will grant much broader protections to expressive initial works. Here, Plaintiff’s copyrighted recordings are inherently expressive and subject to greater protection. Copyrighted recordings are undoubtedly the type of “creative expression for public dissemination [that] falls within the core of the copyright’s protective purposes.” Id.

Importantly, this factor would be mitigated if the Court found that Suno merely used functional portions of Plaintiff’s work. Courts afford less protection to works (or portions of works) that are functional rather than expressive. In Oracle, Sega, Google the Court found “back-end” use of code and written works were fair use because the derivative use of the copied works was merely functional and the use of such works was a technical necessity to achieve the derivative work. Here, Suno argues that their nature of use was functional because their model simply “derives statistical insights about those recordings” to support their AI’s function. Suno will have a harder time proving that their use was a technical necessity. Similar models could have been trained on works in the public domain, for instance. The expressive nature of an AI model’s use of copyrighted recordings is a novel issue to the court, and an important one. If the Court agrees with Suno that their use merely derives statistical insights, not expressive ones, then they’d be more likely to find in favor of Suno on Fair Use.

Factor Three: The Amount of The Use

The third factor weighs against a finding of Fair Use because Suno likely used the entirety, or the key parts, of Plaintiff’s works. The third factor weighs the amount of the original work used in the creation of the derivative work. A derivative work is more likely to be fair use when it uses a smaller portion of the initial work. A derivative work that uses “heart” of the initial work is subject to greater scrutiny. Here, UMG provides evidence of AI outputs replicating key expressive elements of the original work, including core melodies and producer tags. This factor weighs against a finding of fair use because Suno’s ingestion of key expressive elements like producer tags and hooks uses a substantial portion of the initial work.

Factor Four: Market Effect

The fourth factor weighs against a finding of Fair Use because Suno’s use directly and unfairly competes with the market value of the copyrighted recordings. The Court is less likely to find fair use where the derivative work directly competes with the initial work for market share. Courts have not definitively grappled with the competition posed by AI trained on musical works, but recent district court decisions demonstrate a willingness to consider market competition heavily in determining fair use outcomes. In Bart, the Court found against the AI’s derivative use specifically for their use of products that they hadn’t licensed. In Kadrey, the Court noted that an argument about the market competition of AI would be compelling, and argued that AI’s derivative use far outpaces anything the court has contemplated.

Here, UMG makes substantive and compelling claims that Suno’s use competes with the original recordings and dramatically outpaces traditional competition. Suno claims that 10,000,000 people have already created musical files using their service. Moreover, UMG argues that Suno’s product poses a threat to the unique market value of each copyrighted work because would-be licensees of proprietary works will simply create AI soundalikes at virtually no cost. Left unchecked, UMG argues, Suno’s product will effectively and prolifically erase the need to license music while profiting through the use of Plaintiff’s recordings.

In response, Suno argues that their use is fair despite the outlined market competition because their use is sufficiently transformative and only uses functional components of the initial work. A derivative work that competes with the original work may nevertheless be fair use when the new work is sufficiently transformative. In Google, the use of copyrighted code to create competing software was fair use despite its competition because the code used was merely “functional” and necessary to create the new work. Similarly In Sega, the use of Sega’s proprietary compatibility software to create new Sega compatible games was fair use despite its competition because the compatibility software was essential functional code needed to effect a new expressive work.

Suno’s defensive argument permitting market competition is not compelling here due to the unique danger of market competition posed by their product. Their cited authorities do not address this unique threat and consider comparatively much smaller dangers. In Sega, the compatibility code used in the derivative work allowed the defendant to pair his existing creative work with Sega’s product. In Google, the code copied was purely functional and amounted to .4% of the total code in question. Importantly, the defendants in both would still need to use creative energy and processes to make new works. This type of use aligns more neatly with the Court’s priority of advancing the sciences and arts. This code did not, however, allow the defendant to create thousands of competing works in mere minutes. This is the type of competition that Suno represents and, by all indications, this is the type of competition that Courts intend to prevent.

Conclusion

In conclusion, the court will likely find that Suno’s "back-end" use of UMG’s copyrighted recordings does not constitute fair use. While Suno’s use of these recordings to train its AI model can be characterized as transformative because it serves a different purpose than the original works and does not reveal the source material to the public, this transformative nature is ultimately outweighed by other factors. Specifically, the inherently expressive nature of the original sound recordings and the substantial use of key elements, such as core melodies and producer tags, weigh heavily against Suno.

Most critically, Suno’s product poses a unique and significant threat of market supplantation. Unlike previous cases where technical copying was found to be fair use, Suno’s technology enables the rapid creation of thousands of competing works that directly vie for royalties with the original recordings. Because Suno's derivative use shares the same commercial purpose and threatens to erase the need to license original music, the court is likely to prioritize protecting the market value of UMG's catalog over the novel technical purpose of Suno's AI model.

References

• 17 U.S.C. § 106    

  
  

• 17 U.S.C. § 107    

  
  

• Answer of Defendant Suno, Inc. to Complaint, UMG Recordings, Inc. v. Suno, Inc., No. 1:24-cv-11611-FDS (D. Mass. Aug. 1, 2024)    

  
  

• Bartz v. Anthropic PBC, No. 3:24-cv-05417, 2025 WL 4810014 (N.D. Cal. June 23, 2025)    

  
  

• Kadrey v. Meta Platforms, Inc., No. 3:23-cv-03417, 2025 WL 4123456 (N.D. Cal. June 25, 2025)    

  
  

• Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994)    

  
  

• Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015)    

  
  

• Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023)    

  
  

• Google LLC v. Oracle Am., Inc., 593 U.S. 1 (2021)    

  
  

• Sega Enters. Ltd. v. Accolade, Inc., 977 F.2d 1510 (9th Cir. 1992)    

  
  

• Complaint, UMG Recordings, Inc. v. Suno, Inc., No. 1:24-cv-11611 (D. Mass. June 24, 2024)