Music Copyright in the Gen AI Age: Where Are We Now?

Imagine you are a musician who has dedicated years of your life creating an album or EP — tinkering with the production, revising lyrics, finding the perfect samples— and now, you have finally shared your art with the world and are thrilled with the project’s success. However, while scrolling on TikTok a few months later, you hear some familiar audio. Wait a minute, is that one of your songs? No… not quite, but why does it sound so similar? Turns out, the song was created using artificial intelligence (“AI”).

Enter Suno and Udio

On June 24, 2024, Universal Music Group (“UMG”), Sony Music Entertainment (“SME”), and Warner Music Group (“WMG”) filed two separate suits against AI-powered music generators, Suno, Inc. (“Suno”) and Uncharted Labs, Inc. (“Udio”), alleging copyright infringement from Suno and Udio’s use of generative AI (“Gen AI”).¹[1]Daniel Tencer, Major Record Companies Sue AI Music Generators Suno, Udio for ‘Mass Infringement’ of Copyright, Music Business Worldwide (June 24, 2024), https://www.musicbusinessworldwide.com/major-record-companies-sue-ai-music-generators-suno-udio-for-mass-infringement-of-copyright/. Gen AI produces new content (e.g., text, images, videos, and sounds) through the use of existing works.²[2]Adam Zewe, Explained: Generative AI, MIT News (Nov. 9, 2023), https://news.mit.edu/2023/explained-generative-ai-1109. Suno and Udio specifically use Gen AI to create “digital music files within seconds of receiving a user’s prompts.”³[3]UMG Recordings, Inc. et al. v. Suno, Inc., No. 1:24-cv-11611, Compl. ¶ 7 (D. Mass. 2024). To achieve this, the companies “train” their large language models (“LLMs”) on incredibly large amounts of copyrighted music to “deduce patterns,” in order to produce content based on the copyrighted material.⁴[4]Id. at ¶ 41. Suno and Udio claim their LLM training is a non-infringing “fair use” of the existing copyrighted music.

In each of these cases, the courts must determine: (1) whether fair use doctrine permits AI companies to use copyrighted material to train their LLMs; and (2) whether AI companies are responsible for AI-generated songs created using copyrighted material on their platforms. Based on prior case law and ongoing litigation decisions, the courts will likely find that both (1) the use of copyrighted music to train LLMs, and (2) AI-generated music produced by Gen AI services, including Suno and Udio, constitute copyright infringement.

A Brief History of Copyright Law and Fair Use

The U.S. Constitution protects copyrighted material to “promote the Progress of Science and useful Arts” by giving authors exclusive rights over their creations.⁵[5]U.S. Const. art. I, § 8, cl. 8. However, to effectively balance innovation with creation, the fair use doctrine permits the limited use of copyrighted material without having to first acquire permission or obtain a license from the copyright owner.⁶[6]17 U.S.C. § 107. To determine whether a taking of copyrighted material is deemed a “fair use,” Congress sets forth the following four factors:

(1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
(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.⁷[7]Id.

The Supreme Court has held that the fourth factor is “undoubtedly the single most important element” to consider in a fair use analysis.⁸[8]See Harper & Row, Publishers, Inc. v. Nation Enterprises, 471 U.S. 539, 566 (1985).

Prior Case Law Involving Fair Use and Gen AI

In recent years, authors and artists have brought numerous copyright infringement claims accusing AI companies of using their copyrighted material to train LLMs without permission. In Bartz v. Anthropic PBC, a group of authors brought an infringement suit against Anthropic, an AI software firm, for using the authors’ books to train LLMs for its AI service, “Claude.”⁹[9]See 787 F.Supp.3d 1007 (N.D. Cal. 2025). Anthropic responded with a fair use defense.¹⁰[10]See Id. at 1018. Focusing mostly on the first fair use factor, and the “input” of copyrighted books for training, a California court held that Anthropic’s use of the copyrighted books was fair use as the use was “exceedingly transformative.”¹¹[11]Id. at 1019. The Supreme Court has held that the purpose and character of a use turns primarily on whether that use adds something new to the original work, “altering the first with new expression, meaning, or message,” thus making it “transformative.”¹²[12]Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 579 (1994). Here, the court reasoned that because Claude “turn[ed] a hard corner and [created] something different” from the original works, the use was transformative.¹³[13]Bartz, 787 F.Supp.3d at 1022. However, the court found that Anthropic’s storage of pirated, third-party copies of books in its training library was infringing. Although the ruling was a major victory for AI training, it also serves as a warning against using pirated data sources.

Similarly, in Kadrey v. Meta Platforms, Inc., another group of authors brought an infringement action against Meta for using their copyrighted works to train LLMs for its AI service, “Llama.”¹⁴[14]See 788 F.Supp.3d 1026 (N.D. Cal. 2025). Meta also responded with a fair use defense.¹⁵[15]Id. The same California court, focusing mainly on the first and fourth factors, yet again held that using copyrighted material to train LLMs was fair use.¹⁶[16]See Id. at 1036. However, despite its decision, the court was wary about a potential misreading. In relation to the outputs produced by Llama, the Kadrey court cautioned other courts against “enabling the creation of a potentially endless stream of competing works” that would cause significant harm to the market.¹⁷[17]Id. While the authors argued that Meta’s use of the copyrighted works harmed the potential market to license their books for LLM training, the court suggested the authors should have instead made the stronger argument that the Llama-generated content was so similar to the original copyrighted works that they “compete with the originals and thereby indirectly substitute for them.”¹⁸[18]Id. at 1057.

What Does This Mean for Music Artists?

Based on these two cases, it appears courts, at least in California, have determined that the first factor tilts in favor of regarding the training of LLMs with copyrighted material as fair use. The results could differ in the Suno and Udio cases, though, if the labels argue that LLM training serves essentially the same purpose as the original copyrighted songs. Whereas Claude and Llama trained their models on copyrighted books for purposes other than reading (i.e. to generate text for users and perform an array of other functions), Suno and Udio train their models simply to generate music for their user listening. Does this not serve the same purpose of listening to the copyrighted works themselves? If the courts adopted this thinking and find the use not transformative, combined with the commercial nature of the use, the first factor would likely weigh in favor of the record labels.

Although the second and third fair use factors were not hugely relevant to the holdings in Bartz and Kadrey, the fourth factor was significant when considering the outputs of Claude and Llama. While both decisions held that the fourth factor favored the AI companies, the Kadrey court’s clarification regarding the potential of market competition leaves this factor uncertain for future cases in the same vein. In the labels’ complaints, they explain that “certain patterns of prompts can cause [Suno and Udio]’s product to generate digital music files that contain melodic and vocal similarities to well-known copyrighted sound recordings.”¹⁹[19]UMG Recordings, Inc. et al. v. Uncharted Labs, Inc. d/b/a Udio.com, No. 1:24-cv-04777, Compl. ¶ 52 (S.D.N.Y. June 24, 2024). If true, the fourth factor could prove incredibly valuable for the labels.

For instance, if the labels can show that enough Suno or Udio users are listening to their own, AI-generated music, instead of the copyrighted sound recordings, the fourth factor would likely tip in favor of the labels. Moreover, the fourth factor would further favor the labels if they can show that some users are not only listening but also distributing their generated outputs to others. The court in Kadrey calls this idea “indirect substitution.”²⁰[20]Kadrey, 788 F. Supp. 3d at 1052. In response, Suno and Udio may argue that their outputs aren’t directly copying any copyrighted material, but direct copying is not always necessary when looking at market harm under the fourth factor.

In their complaint against Udio, the labels demonstrated the possibility of market substitution by prompting the AI service to produce an audio sample that sounded similar to the classic song, “My Girl” by The Temptations.²¹[21]UMG v. Uncharted Labs, Compl. ¶ 53. On three separate occasions the labels received an output that “any listener familiar with The Temptations would instantly recognize as resembling… ‘My Girl’ (a UMG-owned copyright).”²²[22]Id. at ¶ 55. The complaint goes on to describe the many similarities between the Udio outputs and “My Girl” including the melody, chords, and backing vocals.²³[23]Id.

However, based on Kadrey, similarity may not be the most pertinent aspect of market substitution when arguing against fair use. As long as Suno and Udio can show that their outputs do not directly infringe on existing copyrighted material (i.e., they do not directly copy the underlying musical composition or sound recording), the fourth factor will likely favor the AI companies. Whereas Llama’s outputs in Kadrey did not generate “more than 50 words from any of the [authors’] books,” the labels can show only that Suno’s and Udio’s outputs, at most, directly copy a lyrical phrase or two, or a short melody line.²⁴[24]Kadrey, 788 F.Supp.3d at 1042; UMG v. Uncharted Labs, Compl. ¶ 55. Thus, “direct substitution” would likely not be a viable argument.

Although direct substitution is a weak argument for the labels, they may argue that AI-generated music serves as an indirect substitute. In Campbell v. Acuff Rose Music, Inc., the Supreme Court held that the only harm that matters under the fourth factor is the “harm of market substitution.”²⁵[25]Campbell v. Acuff-Rose Music, Inc., 510 U.S. 593. Therefore, the labels only need to show that Suno and Udio’s outputs, to some degree, displace the copyrighted works in the marketplace.

Where should the industry go from here?

On October 29, 2025 and November 19, 2025, respectively, UMG and WMG both reached compensatory settlements with Udio in the form of a new AI service that seemingly incorporates some sort of blanket license, in which Udio will pay a fee for permission to use all works owned by UMG and WMG to train their models.²⁶[26]Universal Music Group, Universal Music Group and Udio Announce Udio’s First Strategic Agreements for New Licensed AI Music‑Creation Platform (October 29, 2025), https://www.universalmusic.com/universal‑music-group-and-udio-announce-udios-first-strategic-agreements-for-new-licensed-ai-music-creation-platform/; Warner Music Group and Udio Collaborate to Build a New Licensed Music Creation Service, Warner Music Group (November 19, 2025), https://www.wmg.com/news/warner-music-group-and-udio-collaborate-to-build-a-new-licensed-music-creation-service. The new platform is expected to be “powered by new cutting-edge generative AI technology that will be trained on authorized and licensed music.”²⁷[27]Id. In the meantime, until the new platform is released, Udio’s existing AI services will remain available to users with amended features, such as fingerprinting and filtering.²⁸[28]Id. Additionally, WMG settled with Suno in the form of a “first-of-its-kind” partnership on November 25, 2025.²⁹[29]Warner Music Group, Warner Music Group and Suno Forge Groundbreaking Partnership, Warner Music Group (Nov. 25, 2025), https://www.wmg.com/news/warner-music-group-and-suno-forge-groundbreaking-partnership. The partnership will involve licensed models and an opt-in option for songwriters and artists for the use of their “name, image, likeness, voice, and compositions in new AI songs.”³⁰[30]Id.

Although the settlements appear to benefit artists, many questions remain unanswered. Will Suno and Udio’s outputs negatively impact the market of original works? If the proverbial cat is out of the bag regarding AI in the music industry, similar to digital streaming in the 2010s, should the major labels simply submit to blanket licensing deals? If so, how would this fee be dispersed among labels and artists? Should more prominent artists be given a larger cut? Would that necessitate the integration of an attribution element? UMG and WMG’s announcements provide many answers, but the future of AI in the music industry is still uncertain and warrants further discussion.

Written by: Sam Woods
Sam is a 2027 J.D. Candidate at Brooklyn Law School.