Trained or Trouble: Is AI Learning or Infringing?

The Rapid Development of Generative AI

            Artificial intelligence threatens to exploit artists and the fruits of their hard-earned success without permission or compensation. Legal questions surrounding the use of generative AI in music creation are rapidly gaining attention in the entertainment industry. Companies such as OpenAI and Anthropic are competitively developing their Large Language Models (LLMs).[1]

Through a process called tokenization, LLMs break down larger texts into smaller, manageable pieces called tokens, such as words or phrases.[2] For example, one token generally equates to four characters in English, and 100 tokens typically amount to 75 words.[3] LLMs use text, punctuation, and special tokens to understand written inputs. Different tokenization techniques split text into words, characters, or subwords and are helpful for sentiment analysis or spelling correction.[4] This technology is advancing rapidly; for instance, in only two months, Anthropic increased the token capacity of its generative AI, “Claude,” so it can process 91,000 more tokens—in one minute.[5]

Label executives are projected to earn up to $130B annually by investing in generative AI to streamline artist management, enhance fan engagement, and optimize marketing strategies.[6] This financial opportunity in the music space is so valuable because improved tokenization techniques and capacities enable AI models to analyze and synthesize the particularities of each song more efficiently, allowing users of generative AI music platforms to create customized tracks thus directly competing with artist’s copyrighted work without compensating them.[7] Conversely, these advancements in generative AI and tokenization capabilities also raise concerns, as it has opened the door for potential misuse, such as misappropriating an artist’s persona through “voice cloning.”[8]

Why the Music Industry is Concerned

             AI voice cloning is a major concern for artists. The process contains six steps: (1) voice sampling, (2) audio analysis, (3) feature extraction, (4) training the LLM, (5) synthesis and fine-tuning, and (6) output generation.[9] Put simply, massive amounts of audio of the artist’s voice are uploaded and then broken down into phonemes (the smallest units of sound in a language) and characteristics like pitch, tone, and speed.[10] From this, the LLM extracts unique features like accent, intonation, and rhythm to train the AI and generate new speech, which becomes synthesized to make the speech sound natural.[11] In music, an artist’s voice is essentially their product, so allowing users to release songs using artists’ voices via generative AI music services creates major concern amongst both the music industry and legal field.[12]

            In 2023, an anonymous TikToker named Ghostwriter977 used AI to create a song titled “Heart on My Sleeve,” using the voices of Drake and The Weeknd.[13] As the first AI-generated song uploaded to streaming services instead of living on social media, it garnered millions of streams across Spotify, TikTok, and YouTube.[14] Over four days, the song received 2,125,000 global streams worth approximately $9,400 in royalties.[15] Since streaming royalties are distributed pro-rata –meaning an overall revenue pool is divided based on the total popularity of tracks– the revenue earned on “Heart on My Sleeve” was money withheld from other artists.[16] Drake and The Weeknd’s label, Universal Music Group (UMG), quickly took down the song, claiming that training generative AI using its artists’ music violates copyright law despite the lack of a valid legal standing.[17]

Last year, Metro Boomin, a four-time Grammy-winning producer, shook the industry by accidentally using an AI-generated sample for his song “BBL Drizzy.”[18] The sample’s vocals, melody, and instrumental were created by King Willonius (a human being) using Udio, a generative AI music creation program.[19] Udio and Suno are AI platforms that allow users to create music by providing prompts, and they have been sued by the Recording Industry Association of America (RIAA) and major record labels like Sony Music, UMG, and Warner Records for alleged mass infringement of copyrighted sound recordings.[20]

Both cases are expected to set a critical legal precedent that will halt market disruption and protect the rights of creators against AI.[21]

Legal Implications 

            There are two layers to the impact of AI on copyright: (1) whether AI-generated material deserves copyright protection and (2) claims copyright owners can assert against AI-generated material.[22] Regarding the first aspect, the U.S. Copyright Office concluded that AI-generated outputs do not contain the human authorship component required for copyright protection.[23] This conclusion gives music rights holders a valid legal method to prevent and combat future “Heart on My Sleeve” predicaments.[24]

The second aspect involves both input and output.[25] Input concerns whether training AI on copyrighted material constitutes infringement or falls under fair use.[26] Output relates to whether AI-generated works, created using copyrighted material, qualify as derivative works—new works based on or incorporating elements of a pre-existing copyrighted work, such as adaptations, translations, or remixes.[27] Without legal direction from the Copyright Office on the issue of training AI on copyrighted material, legislative action may be the necessary avenue for artists to gain protection from exploitation of their work.[28]

Legislative Protection Against AI Exploitation 

            Several pieces of legislation were introduced in 2024 to combat AI’s unauthorized use of voice and visual likeness. The ELVIS Act was signed into law in Tennessee on March 21, 2024, and the NO FAKES Act was introduced to the U.S. Senate on July 31, 2024.[29] The Generative AI Disclosure Act of 2024 was introduced to the U.S. House of Representatives on April 9, 2024[30]. It would require submission of “a sufficiently detailed summary” of copyrighted works used in training datasets to the Register of Copyrights.[31]

On January 29, 2025, the U.S. Copyright Office released part II of a report discussing the copyrightability of outputs created using generative AI.[32] It concluded that there are circumstances where AI-generated outputs reflect “sufficient human contribution to warrant copyright protection,” but “prompts alone…are unlikely to satisfy those requirements.”[33] Part III will address the legal implications of training AI models on copyrighted works and licensing considerations.[34]

Case Study: UMG Recordings, Inc. v. Uncharted Labs, Inc.

             All major players are holding their breath, waiting to see how UMG v. Uncharted Labs, Inc. d/b/a Udio is decided. The lawsuit was brought on June 6, 2024 and is the first case to be brought by recorded music rights holders against AI. It will also be the first to address the legality of training AI models using copyright-protected music. The major record labels sued Udio for copyright infringement based on the company’s alleged direct copying of the labels’ copyrighted sound recordings to train its AI program that allows users to create music using the copyrighted music of UMG artists as foundational tools.[35]   

Udio asserts a fair use defense, which promotes “freedom of expression by allowing the unauthorized use of copyright-protected works under certain circumstances.” [36] Courts evaluate fair use by considering (1) the purpose and character of the use, (2) the nature of the copyrighted work, (3) the amount and substantiality of the portion taken, and (4) the effect of the use upon the potential market.[37] The outcome of this case will most likely hinge on the first two factors—specifically, whether the use is transformative and adds something new with different character, and how significant the financial impact is on the copyright owner. The court will assess the commercial aspect of Udio’s unlicensed use and examine whether it unfairly competes with the record labels’ access to “both existing and potential commercial markets for selling, licensing, and distributing sound recordings.”[38]

Regarding the first factor, Udio cited multitudes of precedents stating that making an intermediate “copy of a protected work as part of back-end technological process” qualifies as fair use. [39] Additionally, the company argued that its service would not impede the labels’ success due to their monopoly in the music industry. The case remains undecided, but the third part of the Copyright Office’s report should help the court address this dispute. Based on copyright law’s historical tendency to adapt to technological advancement, the labels may be fighting a losing battle. In the 1960s, leaders of the American Federation of Musicians passed a resolution banning synthesizers, fearing they would replace instrumentalists.[40] In the 1980s, sampling was met with intense backlash and was seen as lazy, but it eventually became the foundation for genres like Hip-Hop and Electronic Dance Music.[41]

What Does the Future Hold?

Accurately predicting the future of generative AI’s place within the music industry is extremely difficult. Years of lawsuits from rampant MP3 file sharing in the late 1990s birthed music streaming services.[42] We may end up with a similarly regulated system agreed upon by labels and AI companies. As of August 2023, UMG and Google are allegedly discussing an agreement allowing users to create “deepfake” songs using AI under a model that would compensate the actual artists.[43] Regardless of the decisions or agreements reached, we can only hope artists will be legally protected from exploitation and fairly compensated for their work.

[1]Michael Chui et al., The Economic Potential of Generative AI: The Next Productivity Frontier, McKinsey Digital (June 14, 2023), https://www.mckinsey.com/capabilities/mckinsey%20digital/our-insights/the-economic-potential-of-generative-ai-the-next-productivity%20frontier#introduction.

[2]What is a Token in AI? Understanding how AI Processes Language with Tokenization, Nebius (February 9, 2025), https://nebius.com/blog/posts/what-is-token-in-ai.

[3]Id.

[4]Id.

[5]Chui, supra note 1.

[6]Id.

[7]Martina, AI vs. Music Industry: The Rise of AI Voice Cloning, iMusician (October 12, 2023), https://imusician.pro/en/resources/blog/the-rise-of-ai-voice-cloning.

[8]Id.

[9]What is Voice Cloning?, IIElevenLabs (January 22, 2024), https://elevenlabs.io/blog/what-is-voice-cloning.

[10]Id.

[11]Sophie Ganion, “The Best Thing I Never Said”: Understanding AI Voice Technology and Its Legal Implications on the Music Industry, Comm/Ent (October 24, 2023), https://uclawsfcomment.org/online-content/understanding-ai-voice-technology#_ftn2.

[12]Martina, supra note 7.

[13]Mia Sato, Drake’s AI Clone is Here–and Drake Might Not Be Able to Stop Him, VERGE (May 1, 2023), https://www.theverge.com/2023/5/1/23703087/ai-drake-the-weeknd-music-copyright-legal-battle-right-of-publicity.

[14]Joe Coscarelli, An A.I. Hit of Fake ‘Drake’ and ‘The Weeknd’ Rattles the Music World, The New York Times (April 24, 2023), https://www.nytimes.com/2023/04/19/arts/music/ai-drake-the-weeknd-fake.html.

[15]Colin Stutz, The Fake Drake AI Song Earned Millions of Streams – But Will Anyone Get Paid?, Billboard (April 19, 2023), https://www.billboard.com/pro/fake-drake-ai-song-earned-millions-streams-get-paid/.

[16]Id.

[17]Archie Brydon, Heart on My Sleeve | AI-generated Drake and Weeknd Track Taken Down After Copyright Claim from Universal Music Group, Whynow (April 18, 2023), https://whynow.co.uk/read/heart-on-my-sleeve-ai-generated-drake-and-weeknd-track-taken-down-after-copyright-claim-from-universal-music-group.

[18]Kristin Robinson, Metro Boomin’s ‘BBL Drizzy’ Is More than A Joke – It Could Signal the Future of Sampling, Billboard (May 15, 2024), https://www.billboard.com/business/tech/metro-boomin-bbl-drizzy-future-ai-sampling-1235682587/.

[19]Id.

[20]AI Music The Next Copyright Claims: Legal Challenges Ahead, Stewart Townsend (November 13, 2024), https://stewarttownsend.com/ai-music-the-next-copyright-claims-legal-challenges-ahead/.

[21]Id.

[22]Rachel Reed, AI Created a Song Mimicking the Work of Drake and The Weeknd. What Does That Mean for Copyright Law?, Harvard Law Today (May 2, 2023), https://hls.harvard.edu/today/ai-created-a-song-mimicking-the-work-of-drake-and-the-weeknd-what-does-that-mean-for-copyright-law/.

[23]U.S. Copyright Office, Copyright and Artificial Intelligence Part 2: Copyrightability (2d ed. 2025).

[24]Id.

[25]Reed, supra note 22.

[26]Id.

[27]Id.

[28]U.S. Copyright Office, supra note 23.

[29]Ashley Harbin, ELVIS Act Becomes Law; Tennessee Safeguards Against AI Deep Fakes, Adams and Reese LLP (April 15, 2024), https://www.adamsandreese.com/news-knowledge/elvis-act-tennessee-safeguards-against-deepfakes; Nina Frazier, NO FAKES Act Introduced In The Senate: Protecting Artists’ Rights In The Age of AI, Recording Academy (August 9, 2024), https://www.recordingacademy.com/advocacy/news/no-fakes-act-introduced-in-the-senate.

[30]Aimee Scala and Hanoch Sheps, Navigating Copyright in AI: The Generative AI Disclosure Act of 2024 and Its Implications, Mazzola Lindstrom LLP (May 20, 2024), https://www.mazzolalindstrom.com/navigating-copyright-in-ai-the-generative-ai-disclosure-act-of-2024-and-its-implications-by-aimee-scala-partner-and-hanoch-sheps-counsel/.

[31]Id.

[32]U.S. Copyright Office, Copyright and Artificial Intelligence, Copyright.gov, https://www.copyright.gov/ai/.

[33]Id.

[34]Id.

[35]UMG Recordings, Inc. v. Uncharted Labs, Inc., 24 Civ 4777 (AKH) (S.D.N.Y 2024).

[36]Id. at 2.

[37]Id. at 6.

[38]Id.

[39]UMG Recordings, Inc., supra note 22, at 34.

[40]Daniel Tencer, As Suno and Udio Admit Training AI with Unlicensed Music, Record Industry Says: ‘There’s Nothing Fair about Stealing an Artist’s Life’s Work.’, Music Business Worldwide (Aug. 5, 2024), https://www.musicbusinessworldwide.com/as-suno-and-udio-admit-training-ai-with-unlicensed-music-record-industry-says-theres-nothing-fair-about-stealing-an-artists-lifes-work/.

[41]Jawara Gordon, The Battle for Perfection: AI vs. Sampling, Medium (Feb. 4, 2023), https://medium.com/@jawaragordon/the-battle-for-perfection-ai-vs-sampling-500e9d773a2b.

[42]Music Industry Updates, The Role of Streaming Services in Reducing Music Piracy: A Comprehensive Analysis, Education (June 2023), https://vocal.media/education/the-role-of-streaming-services-in-reducing-music-piracy-a-comprehensive-analysis.

[43]Jake Terrell, How Songwriters, Artists and Labels are Evolving in the AI Era, BENlabs (Aug. 21, 2023), https://www.benlabs.com/resources/ai-in-music-industry/.