Copyrightability and Infringement: A Look at Responses to Artificial Intelligence

Recent advances in generative artificial intelligence (AI) threaten to transform the entertainment industry. As generative AI tools become embedded in everyday creative workflows, fundamental questions about copyright ownership, the legality of using copyrighted works to train AI models and adequacy of existing laws to protect artists have moved to the forefront of the legal discourse. This alert examines three central issues: 1) whether AI-generated works can receive copyright protection, 2) the use of copyrighted materials – particularly music – to train AI models, and 3) the wave of new state and proposed federal legislation designed to protect artists from unauthorized AI exploitation.

Copyrightability of AI-Generated Works

The Human Authorship Requirement

The current position of the U.S. Copyright Office (USCO) is that works generated wholly by AI cannot be registered for copyright protection. The position was conveyed in USCO's letters regarding the case of Kristina Kashtanova, who, on September 15, 2022, submitted an application to copyright a graphic novel called Zarya of the Dawn.¹ The graphic elements of work had been produced using the generative AI service Midjourney,² which led USCO to find the work did not contain "enough original human authorship to sustain a claim to copyright."³ Though Kashtanova argued that she had guided the AI generation process through her "use of textual prompts," USCO rejected this argument.⁴

The USCO's policy statement is rooted in a history of courts finding human involvement central to the concept of authorship as it exists in copyright law, such as the 2018 U.S. Court of Appeals for the Ninth Circuit case Naruto v. Slater, which held that "animals other than humans […] lack statutory standing to sue under the Copyright Act."⁵ USCO made the human authorship requirement explicit in its 2021 Compendium of U.S. Copyright Officer Practices, which includes in its list of uncopyrightable works a "photograph taken by a monkey" and "a song naming the Holy Spirit as the author of the work."⁶

Despite the human authorship requirement, the law makes clear that human "arrangement" of a non-human work or "a sufficient human nexus" can establish the human authorship of a non-human work, and such works are copyrightable.⁷

Thaler v. Perlmutter: The Landmark Case

The seminal decision in Thaler v. Perlmutter addressed whether a work generated entirely by an AI system – with no human involvement – could be copyrighted.⁸ Dr. Stephen Thaler listed his "Creativity Machine" as the sole author of a visual artwork piece titled "A Recent Entrance to Paradise" and sought to claim the copyright himself under a work-made-for-hire theory.⁹ The district court held that human authorship is an essential part of a valid copyright claim and upheld USCO's denial of registration.¹⁰ The U.S. Court of Appeals for the District of Columbia Circuit affirmed in March 2025, holding that an AI computer system is not human and, therefore, "could not register copyright in work it generated."¹¹ Critically, the appellate court clarified that the human authorship requirement "does not prohibit copyrighting work that was made by or with the assistance of artificial intelligence" – it simply requires that the author of that work be a human being, "not the machine itself."¹² The court also noted that Dr. Thaler failed to explain how a ban on machines being authors would disincentivize creativity, because "machines, including the Creativity Machine, do not respond to economic incentives."¹³ Dr. Thaler filed a petition for certiorari, arguing that the Copyright Act's text does not require traditional human contribution and that denying copyright to AI-generated works would discourage investment in AI development and harm creative industries.¹⁴

The Spectrum of Human Involvement: When Does AI-Generated Work Qualify?

In its Kashtanova decision, USCO noted that "[w]hile additional prompts applied to one of these initial images can influence the subsequent images, the process is not controlled by the user because it is not possible to predict what Midjourney will create ahead of time" and "because the software's output was impossible to predict, the software was 'different for copyright purposes than other tools used by artists.'"¹⁵ USCO's reasoning appears to convey that if the intermediary's output cannot be predicted based on certain human inputs, human involvement does not exercise control over the result, and the human, therefore, is not authoring the work.

In Thaler, the D.C. Circuit provided no additional clarity on the "line-drawing disagreements over how much artificial intelligence contributed to a particular human author's work," an endeavor the court dismissed as "neither here nor there in this case."¹⁶ Instead, the court focused narrowly on AI as the sole author of a work, which it found non-copyrightable, sidestepping entirely the issue of how much human involvement would constitute authorship.¹⁷ The courts, therefore, have not yet provided a brightline test for the degree of human involvement necessary to establish human authorship in a non-human work.

Proposed Legal Frameworks: Work-Made-for-Hire and Legislative Reform

Rather than reforming the human authorship requirement, authors Giovanni LoMonaco and Gia Jung point to the works-made-for-hire doctrine as an existing avenue for protecting non-human authorship. LoMonaco cites works-made-for-hire as an exception to the general statutory rule "that copyright ownership 'vests initially in the author or authors of the work.'"¹⁸ Instead, "If a work is made for hire, then 'the employer or other person for whom the work was prepared is considered the author,' and initially owns the copyright," and "when the human 'employer' commissions the AI to create the work, once it begins […] it is 'working within the scope of its employment.'"¹⁹

Jung makes the case that because "corporations are non-human entities in which copyright vests" via works-made-for-hire, this legal concept demonstrates that "copyright can subsist in a non-human work."²⁰ The Thaler district court, however, rejected LoMonaco's and Jung's arguments on this point, holding that because copyright requires a human author, which is absent in the case of AI, "the work at issue did not give rise to a valid copyright upon its creation"; therefore, "property transfer cannot be implicated where no property right exists to transfer in the first instance."²¹

Ultimately, both LoMonaco and Jung seem to concede that the current works-made-for-hire doctrine does not automatically encompass their contemplated paradigm of AI authorship vesting in human (or corporate) employers, and so both call for amending Section 101 of the Copyright Act to include works generated by computers without human authors in the definition of an "employee" for the purposes of works-made-for-hire.²²

AI Use of Copyrighted Music and Materials for Training and Generation

How AI Works: Data Ingestion and the Input Copyright Question

Generative AI systems are trained on vast datasets assembled by scraping content from the internet – including copyrighted works.²³ This training process involves making digital copies of existing works, which AI companies acknowledge.²⁴ The music industry has been particularly impacted, as generative AI programs can now produce content that used to require human input and labor.²⁵ The "Heart on My Sleeve" incident, an AI-generated song mimicking the voices of Drake and The Weeknd that went viral before being taken down, epitomized the threat.²⁶

The Fair Use Debate

Opposite the issue of whether the outputs of an AI system are copyrightable lies the issue of whether inputting copyrighted material to train AI constitutes copyright infringement. AI large language models use machine learning algorithms to analyze "large datasets, like catalogs of music" and generate "an output based on the statistical patterns in the dataset."²⁷

The paramount legal question is whether AI's use of copyrighted material in its training dataset "infringes on copyright by feeding" such material "into an AI program without the consent of the copyright owner during the training process." Both sides of this issue have contended with it by addressing the extent to which such training falls under the purview of fair use under the Copyright Act.

Brooke Sause outlines four considerations for determining whether AI's use of copyrighted material in its training dataset constitutes infringement or fair use:

(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.²⁸

Sause focuses her analysis on the final prong, examining the effect of AI use on the market for the copyrighted work. Here, she points to The Authors Guild, Inc. v. Google, Inc., which found that "Google's unauthorized digitizing of copyright-protected works, creation of a search functionality, and display of snippets from those works are non-infringing fair uses" because "purpose of the copying is highly transformative, the public display of text is limited, and the revelations do not provide a significant market substitute for the protected aspects of the originals."²⁹ Thus, Sause's argument rests on whether AI training creates a market substitute for the work used in that training.³⁰

OpenAI, conversely, argues that AI's use constitutes fair use because "the purpose served by the systems differs from that of the original copyrighted works."³¹ Though "the purpose of the original copyrighted work 'is the direct human consumption of the author's expression,'" OpenAI's purpose in copying the works is "for computer programs to 'learn the patterns inherent in human generated media.'"³² Rather than a market substitute, OpenAI contends its ultimate output is "to create a useful generative AI system rather than [to] generate an expressive work for human consumption."³³

When tested in the courts, the argument that AI training constitutes copyright infringement has not prevailed. The U.S. District Court for the Northern District of California in Kadrey v. Meta Platforms, Inc. did not find infringement because the AI in this case, called Llama, did not reproduce "significant portions of the plaintiffs' books," which would have threatened the market for those books because people could have read the AI outputs instead; here, the AI did "not allow users to generate any meaningful portion of the plaintiffs' books."³⁴ Thus, despite growing use of AI-generated output as a market substitute for copyrighted works, the court found that the use of copyrighted materials in AI training did not alone make the output an infringement of the training material's copyright.

Key Litigation: Concord Music Group v. Anthropic

Concord Music Group v. Anthropic represents the highest-profile case testing AI training practices in the music context. Eight music publishers sued Anthropic, alleging that Anthropic used their copyrighted song lyrics to train its Claude AI model. In denying a preliminary injunction, the Northern District of California pointed to the emerging market for licensing to AI developers, using this market to illustrate that the value of any potential loss "could be ascertained" and, therefore, that no "irreparable harm" existed.³⁵

The Emerging Licensing Market

Though the court in Concord Music Group points to an emerging market for licensing copyrighted material to AI services and Universal Music Group (UMG) "has claimed to be open to exploring licensing content to generative AI companies to build a fair and lawful business model," AI developers have less incentive to license when in cases such as Kadrey and Concord, the courts have not found copyright infringement.³⁶ Likely for this reason, UMG has suggested amending Section 106 of the Copyright Act to make "the use of copyrighted works for training of generative AI […] an exclusive right of the copyright owner," either "as a new right or added as an explicit violation of the reproduction right."³⁷

Nevertheless, some artists have already entered AI-related partnerships. The singer Grimes "collaborated with an AI company to reveal software that allows creators to form songs using her voice," with "any for-profit use of this software" requiring that royalties be split.³⁸ Absent a statutory change, however, the courts have so far avoided finding copyright infringement in cases where AI merely trained on copyrighted materials, and partnerships such as Grimes' will likely depend on AI companies' willingness to entertain them.

Case Study: Timbaland and Suno AI

Recent litigation and public controversy surrounding Grammy‑winning producer Timbaland and the generative music platform Suno Inc. (Suno) provide a concrete, real‑world illustration of the legal issues discussed above – particularly the use of copyrighted recordings in AI training and the limits of fair use. Beginning in 2024, major record labels, through the Recording Industry Association of America (RIAA), filed federal copyright infringement lawsuits against Suno (and a competing platform, Udio) alleging that the companies had copied vast quantities of copyrighted sound recordings without authorization to train their generative AI models.

The controversy intensified in 2025 when Timbaland publicly demonstrated an AI‑generated artist named TaTa created using Suno. Critics alleged the AI outputs bore striking similarities to existing works, including identifiable beats and producer tags, raising concerns that copyrighted recordings may have been ingested during training or reproduced in outputs without consent or attribution. Although Timbaland characterized the demonstration as experimental and non‑commercial, the incident became emblematic of broader industry tensions over transparency, consent and compensation in AI music generation.

Independent artists have also brought lawsuits alleging that Suno's training practices and outputs unlawfully exploit copyrighted works. Suno has countered that its models do not reproduce "samples" of sound recordings and that any learning from copyrighted inputs constitutes non‑infringing fair use.

These Suno cases illustrate the evidentiary and doctrinal difficulties of applying traditional copyright tests – such as substantial similarity – to AI outputs, particularly where AI systems function as "black boxes" and cannot readily disclose the specific works used in training. In this context, the presence of recognizable stylistic elements, producer tags or vocal qualities have become central to plaintiffs' arguments. Moreover, the controversy underscores the growing gap between existing copyright law – developed for human authorship and discrete acts of copying – and generative AI systems that rely on large‑scale data ingestion and probabilistic output generation.

Industry Response: Concerns from Broadcast Music Inc. (BMI) and Major Record Labels

Performance rights organizations and major music companies have taken increasingly public positions in response to these developments. BMI has explicitly rejected claims that training AI models on copyrighted music without permission constitutes fair use, characterizing such practices as unauthorized exploitation that undermines songwriter and composer livelihoods. Though BMI supports the use of AI as a creative tool where human authorship remains central, it has drawn a firm line against unlicensed AI training on protected works, calling instead for consent, compensation and transparency.

Similarly, some major record labels – Universal Music Group, Sony Music Entertainment and Warner Music Group – have expressed concern that generative AI systems trained on copyrighted recordings threaten revenue loss, market substitution and brand dilution. The lawsuits against Suno and Udio reflect a broader strategy to force either judicial recognition of infringement or the creation of mandatory licensing frameworks for AI training datasets.

Collectively, these industry responses reinforce that absent clear statutory reform, courts will continue to resolve AI copyright disputes on a case‑by‑case basis, leaving artists, rightsholders and AI developers operating in a legally uncertain environment.

State and Federal Legislation Protecting Artists from AI

Rather than employing copyright law to protect artists from the potential market effects of AI, several states have instead turned to the right of publicity. The right of publicity statutes in Illinois, Utah, Arkansas and Montana include explicit reference to AI or generative AI. California, Illinois and Kentucky have proposed legislation to regulate "digital replicas" of voices or likenesses, and Louisiana has proposed legislation to regulate "deepfakes."³⁹

Tennessee's Ensuring Likeness Voice and Image Security (ELVIS) Act more broadly regulates any "simulation of the voice of the individual" rather than limiting the restriction to digital or AI recreations, which has resulted in the John R. Cash Revocable Trust suing The Coca-Cola Company (TCCC) under the ELVIS Act over what the former alleges was TCCC's use in a television commercial not of an AI deepfake but of "a Johnny Cash tribute singer […] to sing the vocal track" of its "Go the Distance" TV commercial.⁴⁰

Efforts to regulate AI to protect artists at the federal level have also focused on the right of publicity approach. The proposed Nurture Originals, Foster Art, and Keep Entertainment Safe (No FAKES) Act "would 'prohibit the unauthorized use of digital replicas without informed consent,'" and the No Artificial Intelligence Fake Replicas and Unauthorized Duplications (No AI FRAUD) Act "would create a federal 'property right' in an individual's 'likeness and voice.'"⁴¹

Key Takeaways and Practical Considerations

Given that courts have not outlined a clear test on the degree of human involvement that would make an AI-generated work copyrightable and have so far not found that the use of copyrighted materials in AI training constitutes infringement, statutory reforms will likely be required to address these legal gaps. Some possible solutions lie in reforming works-made-for-hire, while others look to the right of publicity rather than copyright law. Several states have already reformed their statutes, and federal legislation on the topic has been proposed.

Nashville songwriter Lee Miller compared the potential effect of AI to the decline of physical media and popularization of streaming in the past decade, noting, "we used to make money on album sales. One million people used to walk into a record store or a Walmart and pay $23 for a CD, and if it went platinum, one million people bought it. We went from that model to zero record sales thanks to tech. Then the industry told consumers music was free, and we lost 30% of income." The AI revolution will likely pose similar challenges and require both legal and market responses to adapt to this new technology. Nevertheless, Miller is optimistic: "Songwriting is not dead and not going to die. The heart of music is in words and stories and nuance. Shifts in music are created by people who have grown up being influenced by music. You can't train a computer to do that."

For more information or questions, please contact the authors.

Notes

¹ Letter from Robert J. Kasunic, Assoc. Reg. of Copyrights and Dir. of the Off. of Registration Pol'y & Prac., U.S. Copyright Off., to Van Lindberg at 1-2 (Feb. 21, 2023).

² Midjourney is a subscription service that allows users to pay to generate images, with subscription plans corresponding to the computational time it uses to generate images. See Id. at 5.

³ Giovanni LoMonaco, "The Times They Are A-Changin': Adapting Copyright Law to Generative AI," 45 Pace L. Rev. 573, 582 (Fall 2024) (citing Letter from Robert J. Kasunic, Assoc. Reg. of Copyrights and Dir. of the Off. of Registration Pol'y & Prac., U.S. Copyright Off., to Kristina Kashtanova at 3 (Oct. 28, 2022)).

⁴ Kasunic to Lindberg, supra note 1, at 9.

⁵ Naruto v. Slater, 888 F.3d 418, 426 (9th Cir. 2018).

⁶ Henry H. Perritt Jr., "Copyright for Robots?," 57 Ind. L. Rev. 139, 168 (2023).

⁷ Perritt, supra note 6, at 159-160; Gia Jung, "Do Androids Dream of Copyright?: Examining AI Copyright Ownership," 35 Berkeley Tech. L.J. 1151, 1162 (2020).

⁸ Thaler v. Perlmutter, 687 F. Supp. 3d 140 (D.D.C. 2023).

⁹ Id. at 143.

¹⁰ Id. at 150.

¹¹ Thaler v. Perlmutter, 130 F.4th 1039, 1039 (D.C. Cir. 2025).

¹² Id. at 1049.

¹³ Id. at 1050.

¹⁴ Thaler v. Perlmutter, 2025 WL 2954208 (2025).

¹⁵ LoMonaco, supra note 3, at 583-584.

¹⁶ Thaler v. Perlmutter, 130 F.4th 1039, 1050 (D.C. Cir. 2025).

¹⁷ Id.

¹⁸ LoMonaco, supra note 3, at 594.

¹⁹ Id. at 594-595.

²⁰ Gia Jung, supra note 7, at 1169-1170.

²¹ Thaler v. Perlmutter, 687 F. Supp. 3d 140, 150 (D.D.C. 2023).

²² LoMonaco, supra note 3, at 597; Jung, supra note 7, at 1173.

²³ Jenna L. Kurtz, "Where Melodies Emerge from Algorithms: A Dissection of the Need for a Licensing Structure in a World of AI-Generated Music," 17 Drexel L. Rev. 241, 250 (2024).

²⁴ Brooke Sause, "The Melodic Maze of Generative AI: Navigating Copyright and Publicity Protections," 4 Fla. Ent. & Sports L. Rev. 107, 110 (February 2025).

²⁵ LoMonaco, supra note 3, at 574.

²⁶ Kurtz, supra note 23, at 244.

²⁷ Id. at 249.

²⁸ Sause, supra note 24, at 110-111.

²⁹ Authors Guild v. Google Inc., 804 F.3d 202, 229 (2d Cir. 2015).

³⁰ Sause, supra note 24, at 114.

³¹ Chanelle Fagbemi, "Resurrected: Music Copyright and Estates in the Age of Generative AI," 16 Am. U. Intell. Prop. Brief 29, 43–45 (June 2025).

³² Fagbemi, supra note 31, at 45.

³³ Id. at 45.

³⁴ Kadrey v. Meta Platforms, Inc., 788 F. Supp. 3d 1026, 1051 (N.D. Cal. 2025).

³⁵ Concord Music Group, Inc. v. Anthropic PBC, 772 F. Supp. 3d 1131, 1134, 1141 (N.D. Cal. 2025).

³⁶ Sause, supra note 24, at 122.

³⁷ Id. at 122.

³⁸ Kurtz, supra note 23, at 257.

³⁹ Sarah Luppen Fowler & John D. Fowler, "Why Tennessee's ELVIS Act Is the King of Artificial Intelligence Protections," 27 Vand. J. Ent. & Tech. L. 259, 270 (Winter 2024).

⁴⁰ Complaint at 9–10, John R. Cash Revocable Trust v. The Coca-Cola Co., No. 3:25-cv-01373 (M.D. Tenn. Nov. 25, 2025).

⁴¹ Id. at 271.

Information contained in this alert is for the general education and knowledge of our readers. It is not designed to be, and should not be used as, the sole source of information when analyzing and resolving a legal problem, and it should not be substituted for legal advice, which relies on a specific factual analysis. Moreover, the laws of each jurisdiction are different and are constantly changing. This information is not intended to create, and receipt of it does not constitute, an attorney-client relationship. If you have specific questions regarding a particular fact situation, we urge you to consult the authors of this publication, your Holland & Knight representative or other competent legal counsel.