Copyright law “has never stretched so far, however, as to protect works generated by new forms of technology operating absent any guiding human hand, as plaintiff urges here. Human authorship is a bedrock requirement of copyright.”
The United States Supreme Court has put an end to Stephen Thaler’s crusade for machine rights. Okay, that’s the sensational news article way of putting it. He wasn’t really crusading for machine rights. He was trying to establish a precedent for claiming copyright in AI-generated works.
I first wrote about this back in May, 2022 (“AI Can Create, But Is It Art?”). At that time, the U.S. Copyright Office had denied registration of “A Recent Entrance to Paradise.” This was an image that was generated by Thaler’s AI tool, the Creativity Machine. Thaler had sought to register it as a work for hire made by the machine. The Copyright Office denied registration because it lacked human authorship.
The decision was consistent with appellate court decisions suggesting that stories allegedly written by “non-human spiritual beings” are not protected by copyright, although a human selection or arrangement of them might be. Urantia Foundation v. Kristen Maaherra, 114 F.3d 955 (9th Cir. 1997). Neither are works created by non-human animals, such as a monkey selfie.
Thaler sought review by the federal district court. Judge Howell affirmed the Copyright Office’s decision, writing that copyright law “has never stretched so far, however, as to protect works generated by new forms of technology operating absent any guiding human hand, as plaintiff urges here. Human authorship is a bedrock requirement of copyright.”
The Court of Appeals affirmed the refusal of registration. Thaler petitioned for review by the United States Supreme Court. On March 2, 2026, the Court denied review, without comment.
An argument that Thaler advanced in the petition for certiorari was bascially that because images output by a camera are protected by copyright (See Burrow-Giles Lithographic v. Sarony), images generated by a computer should be, too.
The Copyright Office has since published guidance explaining that using AI as a tool in the creative process does not categorically rule out copyright protection. Rather, assessments must be made on a case-by-case basis about the nature and extent of human creativity that was contributed.
The narrowest interpretation of the Supreme Court’s denial of certiorari is that it did not see a need to disturb the ruling that a machine cannot be an “author,” for purposes of copyright law. The facts of the case did not present an opportunity to opine on whether, and under what circumstances, a human can claim to be an author of an AI-assisted creation.
Trademark News
Buc-ee’s, a popular chain of gas-and-convenience stores in the South, has filed a trademark infringement lawsuit against Mickey’s gas stations. According to the complaint:
Consumers are likely to perceive a connection or association as to the source, sponsorship, or affiliation of the parties’ products and services, when in fact none exists, given the similarity of the parties’ logos, trade channels, and consumer bases.
Here are the two logos, side by side for comparison:
Trademark infringement occurs when one company’s logo or other mark is used in commerce in a way that is likely to confuse consumers about the source of a product or service. What do you think, folks? Might a weary traveler mistake a moose for a beaver?
News media headlines are trumpeting that the Executive Order preempts state AI laws. This is not true. It directs this administration to try to strike down some state AI laws. It contemplates working with Congress to formulate and enact preemptive legislation. It is doubtful that a President could constitutionally preempt state laws by executive order.
On December 11, 2025, President Trump issued another Executive Order. This one is intended to promote “national dominance” in “a race with adversaries for supremacy.” To “win,” the Order says, AI companies should not be encumbered by state regulation. “The policy of the United States,” the Order says, is “to sustain and enhance the United States’ global AI dominance through a minimally burdensome national policy framework for AI.” It sets up an AI Litigation Task Force to challenge state AI laws that allegedly do not do that.
Excepted from the Order are state laws on child safety protections, data center infrastructure, and state government use of AI.
Which State AI Laws?
The Order speaks generally about “state AI laws,” but does not define the term. Here are some examples of state AI laws:
Stalking and Harassment
A North Dakota statute criminalizes using a robot to frighten or harass another person. It defines a robot to include a drone or other system that uses AI technology. (N.D. Cent. Code § 12.1-17-07.(1), (2)(f)). This appears to be a “state AI law.” North Dakota statutes also prohibit stalking accomplished by using either a robot or a non-AI form of technology. (N.D. Cent. Code § 12.1-17-07.1(1)(d)). Preempting this statute would produce an anomalous result. It would be a crime to stalk somebody unless you use an AI-powered device to do it.
Political Deepfakes
Several states have enacted laws prohibiting the distribution of political deepfakes to influence an election. Regulations range from a prohibition against the distribution of a deepfake to influence an election within a specified time period before the election to requiring disclosure that it is AI-generated. Minn. Stat. § 609.771 is an example of such a regulation. The need for this kind of statute was highlighted in 2024 when someone used AI to clone Joe Biden’s voice and generate an audio file that sounded like Mr. Biden himself was urging people not to vote for him.
Sexual Deepfakes
Both state and federal governments have enacted laws aimed at curbing the proliferation of “revenge porn.” The TAKE IT DOWN Act is an example. Minn. Stat. § 604.32 is another example (deepfakes depicting intimate body parts or sexual acts).
State and federal laws in this area cover much of the same ground. The principal difference is that the federal crime must involve interstate commerce; state crimes do not. The only practical effect of preemption of this kind of state AI law, therefore, would be to eliminate state prohibitions of wholly intrastate sexual deepfakes. If the Executive Order succeeds in its objectives, then state laws that prohibit the creation or distribution of sexual deepfakes wholly within the same state, as some do, would be preempted, with the result that making and distributing sexual deepfakes would be lawful so long as you only transmit it to other people in your state and not to someone in a different state.
Digital Replicas
Many states have enacted laws prohibiting or regulating the unauthorized creation and exploitation of digital replicas. The California Digital Replicas Act and Tennessee’s ELVIS Act are examples. AI is used in the creation of digital replicas. It is unclear whether these kinds of enactments are “state AI laws.” Arguably, a person could use technologies more primitive than generative-AI to create a digital image of a person. If these statutes are preempted only to the extent they apply to AI-generated digital replicas, then it would seem that unauthorized exploiters of other people’s faces and voices for commercial gain would be incentivized to use AI to engage in unauthorized commerceial exploitation of other people.
Child Pornography
Several states have either enacted laws or amended existing laws to bring AI-generated images of what look like real children within the prohibition against child pornography. See, e.g., N.D. Cent. Code § 12.1.-27.2—01. The Executive Order exempts “child safety protections,” but real children do not necessarily have to be used in AI-generated images. This kind of state statute arguably would not come within the meaning of a “child safety protection.”
Health Care Oversight
California’s Physicians Make Decisions Act requires a human person to oversee health care decisions about medical necessity. This is to ensure that medical care is not left entirely up to an AI bot. The law was enacted with the support of the California Medical Association to ensure that patients receive adequate health care. If the law is nullified, then it would seem that hospitals would be free to replace doctors with AI chatbots.
Chatbots
Some states prohibit the deceptive use of a chatbot, such as by falsely representing to people who interact with one that they are interacting with a real person. In addition, some states have enacted laws requiring disclosure to consumers when they are interacting with a non-human AI. See, e.g., the Colorado Artificial Intelligence Act.
Privacy
Some states have enacted either stand-alone laws or amended existing privacy laws to ensure they protect the privacy of personally identifiable information stored by AI systems. See, e.g., Utah Code 13-721-201, -203 (regulating the sharing of a person’s mental health information by a chatbot); and amendments to the California Consumer Privacy Act making it applicable to information stored in an AI system.
The Texas Responsible Artificial Intelligence Governance Act
Among other things, the Texas Responsible AI Governance Act prohibits the use of AI to restrict constitutional rights, to discriminate on the basis of race, or to encourage criminal activity. These seem like reasonable proscriptions.
Trump’s “AI czar,” venture capitalist David Sacks, has said the administration is not gong to “push back” on all state laws, only “the most onerous” ones. It is unclear which of these will be deemed “onerous.”
State AI Laws are Not Preempted
News media headlines are trumpeting that the Executive Order preempts state AI laws. This is not true. It directs this administration to try to strike down some state AI laws. It contemplates working with Congress to formulate and enact preemptive legislation. It is doubtful that a President could constitutionally preempt state laws by executive order.
Postscript
Striving for uniformity in the regulation of artificial intelligence is not a bad idea. There should be room, though, for both federal and state legislation. Rather than abolishing state laws, a uniform code or model act for states might be a better idea. Moreover, if we are going to start caring about an onerous complex of differing state laws, and feeling a need to establish a national framework, perhaps the President and Congress might wish to address the sprawling morass of privacy and data security regulations in the United States.
Copyright cannot be claimed in a voice. Copyright law protects only expression, not a person’s corporeal attributes.
Painting of Nipper by Francis Barraud (1898-99); subsequently used as a trademark with “HIs Master’s Voice.”
Lehrman v. Lovo, Inc.
On July 10, 2025, the federal district court for the Southern District of New York issued an Order granting in part and denying in part a motion to dismiss a putative class action lawsuit that Paul Lehrman and Linnea Sage commenced against Lovo, Inc. The lawsuit, Lehrman v. Lovo, Inc., alleges that Lovo used artificial intelligence to make and sell unauthorized “clones” of their voices.
Specifically, the complaint alleges that the plaintiffs are voice-over actors. For a fee, they read and record scripts for their clients. Lovo allegedly sells a text-to-speech subscription service that allows clients to generate voice-over narrations. The service is described as one that uses “AI-driven software known as ‘Generator’ or ‘Genny,'” which was “created using ‘1000s of voices.'” Genny allegedly creates voice clones, i.e., copies of real people’s voices. Lovo allegedly granted its customers “commercial rights for all content generated,” including “any monetized, business-related uses such as videos, audio books, advertising promotion, web page vlogging, or product integration.” (Lovo terms of service.) The complaint alleges that Lovo hired the plaintiffs to provide voice recordings for “research purposes only,” but that Lovo proceeded to exploit them commercially by licensing their use to Lovo subscribers.
This lawsuit ensued.
The complaint sets out claims for:
Copyright infringement
Trademark infringement
Breach of contract
Fraud
Conversion
Unjust enrichment
Unfair competition
New York civil rights laws
New York consumer protection laws.
The defendant moved to dismiss the complaint for failure to state a claim.
The copyright claims
Sage alleged that Lovo infringed the copyright in one of her voice recordings by reproducing it in presentations and YouTube videos. The court allowed this claim to proceed.
Plaintiffs also claimed that Lovo’s unauthorized use of their voice recordings in training its generative-AI product infringed their copyrights in the sound recordings. The court ruled that the complaint did not contain enough factual detail about how the training process infringed one of the exclusive rights of copyright ownership. Therefore, it dismissed this claim with leave to amend.
The court dismissed the plaintiffs’ claims of output infringement, i.e., claims that the “cloned” voices the AI tool generated infringed copyrights in the original sound recordings.
Copyright protection in a sound recording extends only to the actual recording itself. Fixation of sounds that imitate or simulate the ones captured in the original recording does not infringe the copyright in the sound recording.
This issue often comes up in connection with copyrights in music recordings. If Chuck Berry writes a song called “Johnny B. Goode” and records himself performing it, he will own two copyrights – one in the musical composition and one in the sound recording. If a second person then records himself performing the same song, and he doesn’t have a license (compulsory or otherwise) to do so, that person would be infringing the copyright in the music but not the copyright in the sound recording. This is true even if he is very good at imitating Berry’s voice and guitar work. For a claim of sound recording infringement to succeed, it must be shown that the actual recording itself was copied.
Plaintiffs did not allege that Lovo used Genny to output AI-generated reproductions of their original recordings. Rather, they alleged that Genny is able to create new recordings that mimic attributes of their voices.
The court added that the sound of a voice is not copyrightable expression, and even if it were, the plaintiffs had registered claims of copyright in their recordings, not in their voices.
The trademark claims
In addition to infringement, the Lanham Act creates two other potential bases of trademark liability: (1) false association; and (2) false advertising. 15 U.S.C. sec. 1125(a)(1)(A) and (B). Plaintiffs asserted both kinds of claims. The judge dismissed these claims.
False association
The Second Circuit court of appeals recently held, in Electra v. 59 Murray Enter., Inc. and Souza v. Exotic Island Enters., Inc., that using a person’s likeness to create an endorsement without the person’s permission can constitute a “false association” violation. In other words, a federally-protected, trademark-like interest in one’s image, likeness, personality and identity exists. (See, e.g., Jackson v. Odenat.)
Although acknowledging that this right extends to one’s voice, the judge ruled that the voices in this case did not function as trademarks. They did not identify the source of a product or service. Rather, they were themselves the product or service. For this reason, the judge ruled that the plaintiffs had failed to show that their voices, as such, are protectable trademarks under Section 43(a)(1)(A) of the Lanham Act.
False Advertising
Section 43(a)(1)(B) of the Lanham Act (codified at 15 U.S.C. sec. 1125(a)(1)(B)) prohibits misrepresentations about “the nature, characteristics, qualities, or geographic origin of . . . goods, services, or commercial activities.” The plaintiffs claimed that Lovo marketed their voices under different names (“Kyle Snow” and “Sally Coleman.”) The court determined that this was not fraudulent, however, because Lovo marketed them as what they were, namely, synthetic clones of the actors’ voices, not as their actual voices.
Plaintiffs also claimed that Lovo’s marketing materials falsely stated that the cloned voices “came with all commercial rights.” They asserted that they had not granted those rights to Lovo. The court ruled, however, that even if Lovo was guilty of misrepresentation, it was not the kind of misrepresentation that comes within Section 43(a)(1)(B), as it did not concern the nature, characteristics, qualities, or geographic origin of the voices.
State law claims
Although the court dismissed the copyright and trademark claims, it allowed some state law claims to proceed. Specifically, the court denied the motion to dismiss claims for breach of contract, violations of sections 50 and 51 of the New York Civil Rights Law, and violations of New York consumer protection law.
Both the common law and the New York Civil Rights Law prohibit the commercial use of a living person’s name, likeness or voice without consent. Known as “misappropriation of personality” or violation of publicity or privacy rights, this is emerging as one of the leading issues in AI law.
The court also allowed state law claims of false advertising and deceptive trade practices to proceed. The New York laws are not subject to the “nature, characteristics, qualities, or geographic origin” limitation set out in Section 43(a) of the Lanham Act.
Conclusion
I expect this case will come to be cited for the rule that copyright cannot be claimed in a voice. Copyright law protects only expression, not a person’s corporeal attributes. The lack of copyright protection for a person’s voice, however, does not mean that voice cloning is “legal.” Depending on the particular facts and circumstances, it may violate one or more other laws.
It also should be noted that after the Joe Biden voice-cloning incident of 2024, states have been enacting statutes regulating the creation and distribution of voice clones. Even where a specific statute is not applicable, though, a broader statute (such as the FTC Act or a similar state law) might cover the situation.
Images and references in this blog post are for illustrative purposes only. No endorsement, sponsorship or affiliation with any person, organization, company, brand, product or service is intended, implied, or exists.
Official portrait of Vice President Joe Biden in his West Wing Office at the White House, Jan. 10, 2013. (Official White House Photo by David Lienemann)
Court rules that using copyrighted works to train AI is fair use. Kadrey et al. v. Meta Platforms.
Just days after the first major fair use ruling in a generative-AI case, a second court has determined that using copyrighted works to train AI is fair use. Kadrey et al. v. Meta Platforms, No. 3:23-cv-03417-VC (N.D. Cal. June 25, 2025).
The Kadrey v. Meta Platforms Lawsuit
I previously wrote about this lawsuit here and here.
Meta Platforms owns and operates social media services including Facebook, Instagram, and WhatsApp. It is also the developer of a large language model (LLM) called “Llama.” One of its releases, Meta AI, is an AI chatbot that utilizes Llama.
To train its AI, Meta obtained data from a wide variety of sources. The company initially pursued licensing deals with book publishers. It turned out, though, that in many cases, individual authors owned the copyrights. Unlike music, no organization handles collective licensing of rights in book content. Meta then downloaded shadow library databases. Instead of licensing works in the databases, Meta decided to just go ahead and use them without securing licenses. To download them more quickly, Meta torrented them using BitTorrent.
Meta trained its AI models to prevent them from “memorizing” and outputting text from the training data, with the result that no more than 50 words and punctuation marks from any given work were reproduced in any given output.
The plaintiffs named in the Complaint are thirteen book authors who have published novels, plays, short stories, memoirs, essays, and nonfiction books. Sarah Silverman, author of The Bedwetter; Junot Diaz, author of The Brief Wondrous Life of Oscar Wao; and Andrew Sean Greer, author of Less, are among the authors named as plaintiffs in the lawsuit. The complaint alleges that Meta downloaded 666 copies of their books without permission and states claims for direct copyright infringement, vicarious copyright infringement, removal of copyright management information in violation of the Digital Millennium Copyright Act (DMCA), and various state law claims. All claims except the ones for direct copyright infringement and violation of the DMCA were dismissed in prior proceedings.
Both sides moved for summary judgment on fair use with respect to the claim that Meta’s use of the copyrighted works to train its AI infringed copyrights. Meta moved for summary judgment on the DMCA claims. Neither side moved for summary judgment on a claim that Meta infringed copyrights by distributing their works (via leeching or seeding).
On June 25, 2025 Judge Chhabria granted Meta’s motion for summary judgment on fair use with respect to AI training; reserved the motion for summary judgment on the DMCA claims for decision in a separate order, and held that the claim of infringing distribution via leeching or seeding “will remain a live issue in the case.”
Judge Chhabria’s Fair Use Analysis
Judge Chhabria analyzed each of the four fair use factors. As is the custom, he treated the first (Character or purpose of the use) and fourth (Effect on the market for the work) factors as the most important of the four.
He disposed of the first factor fairly easily, as Judge Alsup did in Bartz v. Anthropic, finding that the use of copyrighted works to train AI is a transformative use. This finding weighs heavily in favor of fair use. The purpose of Meta’s AI tools is not to generate books for people to read. Indeed, in this case, Meta had installed guardrails to prevent the tools from generating duplicates or near-duplicates of the books on which the AI was trained. Moreover, even if it could allow a user to prompt the creation of a book “in the style of” a specified author, there was no evidence that it could produce an identical work or a work that was substantially similar to one on which it had been trained. And writing styles are not copyrightable.
Significantly, the judge held that the use of shadow libraries to obtain unauthorized copies of books does not necessarily destroy a fair use defense. When the ultimate use to be made of a work is transformative, the downloading of books to further that use is also transformative, the judge wrote. This ruling contrasts with other judges who have intimated that using pirated copies of works weighs against, or may even prevent, a finding of fair use.
Unlike some judges, who tend to consider the fair use analysis over and done if transformative use is found, Judge Chhabria recognized that even if the purpose of the use is transformative, its effect on the market for the infringed work still has to be considered.
3 Ways of Proving Adverse Market Effect
The Order lays out three potential kinds of arguments that may be advanced to establish the adverse effect of an infringing use on the market for the work:
The infringing work creates a market substitute for the work;
Use of the work to train AI without permission deprives copyright owners of a market for licenses to use their works in AI training;
Dilution of the market with competing works.
Market Substitution
In this case, direct market substitution could not be established because Meta had installed guardrails that prevented users from generating copies of works that had been used in the training. Its AI tools were incapable of generating copies of the work that could serve as substitutes for the authors’ works.
The Market for AI Licenses
The court refused to recognize the loss of potential profits from licensing the use of a work for AI training purposes as a cognizable harm.
Market Dilution
The argument here would be that the generation of many works that compete in the same market as the original work on which the AI was trained dilutes the market for the original work. Judge Chhabria described this as indirect market substitution.
The copyright owners in this case, however, focused on the first two arguments. They did not present evidence that Meta’a AI tools were capable of generating books; that they do, in fact, generate books; or that the books they generate or are capable of generating compete with books these authors wrote. There was no evidence of diminished sales of their books.
Market harm cannot be assumed when generated copies are not copies that can serve as substitutes for the specific books claimed to have been infringed. When the output is transformative, as it was in this case, market substitution is not self-evident.
Judge Chhabria chided the plaintiffs for making only a “half-hearted argument” of a significant threat of market harm. He wrote that they presented “no meaningful evidence on market dilution at all.”
Consequently, he ruled that the fourth fair use factor favored Meta.
Conclusion
The decision in this case is as significant for what the court didn’t do as it is for what it did. It handed a fair use victory to Meta. At the same time, though, it did not rule out a finding that training AI tools on copyrighted works is not fair use in an appropriate case. The court left open the possibility that a copyright owner might prevail on a claim that training AI on copyrighted works is not fair use in a different case. And it pointed the way, albeit in dictum, namely, by making a strong showing of market dilution.
Anthropic also acquired infringing copies of works from pirate sites. Judge Alsup ruled that these, and uses made from them, are not fair use.
A federal judge has issued a landmark fair use decision in a generative-AI copyright infringement lawsuit.
In a previous blog post, I wrote about the fair use decision in Thomson Reuters v. ROSS. As I explained there, that case involved a search-and-retrieval AI system, so the holding was not determinative of fair use in the context of generative AI. Now we finally have a decision that addresses fair use in the generative-AI context.
Bartz et al. v. Anthropic PBC
Anthropic is an AI software firm founded by former OpenAI employees. It offers a generative-AI tool called Claude. Like other generative-AI tools, Claude mimics human conversational skills. When a user enters a text prompt, Claude will generate a response that is very much like one a human being might make (except it is sometimes more knowledgeable.) It is able to do this by using large language models (LLMs) that have been trained on millions of books and texts.
Adrea Bartz, Charles Graeber, and Kirk Wallace Johnson are book authors. In August 2024, they sued Anthropic, claiming the company infringed the copyrights in their works. Specifically, they alleged that Anthropic copied their works from pirated and purchased sources, digitized print versions, assembled them into a central library, and used the library to train LLMs, all without permission. Anthropic asserted, among other things, a fair use defense.
Earlier this year, Anthropic filed a motion for summary judgment on the question of fair use.
On June 23, 2025, Judge Alsup issued an Order granting summary judgment in part and denying it in part. It is the first major ruling on fair use in the dozens of generative-AI copyright infringement lawsuits that are currently pending in federal courts.
The Order includes several key rulings.
Digitization
Anthropic acquired both pirated and lawfully purchased printed copies of copyright-protected works and digitized them to create a central e-library. Authors claimed that making digital copies of their works infringed the exclusive right of copyright owners to reproduce their works. (See 17 U.S.C. 106.)
In the process of scanning print books to create digital versions of them, the print copies were destroyed. Book bindings were stripped so that each individual page could be scanned. The print copies were then discarded. The digital copies were not distributed to others. Under these circumstances, the court ruled that making digital versions of print books is fair use.
The court likened format to a frame around a work, as distinguished from the work itself. As such, a digital version is not a new derivative work. Rather, it is a transformative use of an existing work. So long as the digital version is merely a substitute for a print version a person has lawfully acquired, and so long as the print version is destroyed and the digital version is not further copied or distributed to others, then digitizing a printed work is fair use. This is consistent with the first sale doctrine (17 U.S.C. 109(a)), which gives the purchaser of a copy of a work a right to dispose of that particular copy as the purchaser sees fit.
In short, the mere conversion of a lawfully acquired print book to a digital file to save space and enable searchability is transformative, and so long as the print version is destroyed and the digital version is not further copied or distributed, it is fair use.
AI Training Is Transformative Fair Use
The authors did not contend that Claude generated infringing output. Instead, they argued that copies of their works were used as inputs to train the AI. The Copyright Act, however, does not prohibit or restrict the reading or analysis of copyrighted works. So long as a copy is lawfully purchased, the owner of the purchased copy can read it and think about it as often as he or she wishes.
[I]f someone were to read all the modern-day classics because of their exceptional expression, memorize them, and then emulate a blend of their best writing, would that violate the Copyright Act? Of course not.
Judge Alsup described AI training as “spectacularly” transformative.” Id. After considering all four fair use factors, he concluded that training AI on lawfully acquired copyright-protected works (as distinguished from the initial acquisition of copies) is fair use.
Pirating Is Not Fair Use
In addition to lawfully purchasing copies of some works, Anthropic also acquired infringing copies of works from pirate sites. Judge Alsup ruled that these, and uses made from them, are not fair use. The case will now proceed to trial on the issue of damages resulting from the infringement.
Conclusion
Each of these rulings seems, well, sort of obvious. It is nice to have the explanations laid out so clearly in one place, though.
Court of Appeals Affirms Registration Refusal for AI-Generated Output
In 2019, Stephen Thaler developed an AI system he called The Creativity Machine. He generated output he called A Recent Entrance to Paradise. When he applied to register a copyright claim in the output, he listed the machine as the author. He claimed ownership of the work as a work made for hire. In his application, he asserted that the work was autonomously created by a machine. The Copyright Office denied the claim on the basis that human authorship is a required element of a copyright claim.
On appeal, the United States district court affirmed the Copyright Office’s decision. Thaler attempted to argue, for the first time, that it was copyrightable because he provided instructions and directed the machine’s creation of the work. The district court found that he had waived that argument.
The Court of Appeals Affirms
Thaler sought review in the Court of Appeals for the Federal Circuit. On March 18, 2025, the Court of Appeals affirmed. The Court cited language in the Copyright Act that suggested Congress intended only human beings to be authors. The Court did not reach the question whether the Copyright Clause of the U.S. Constitution might protect machine-generated works if Congress should choose someday to extend copyright protection to these kinds of materials.
The Court held that the question whether Thalercould claim authorship on the basis of the fact that he made and directed the operation of the Creativity Machine has not been preserved for appeal.
Copyright registrations are being issued for works created with generative-AI tools, subject to some important qualifications. Also, Internet Archves revisited (briefly)
The U.S. Copyright Office has issued its long-awaited report on the copyrightability of works created using AI-generated output. The legality of using copyrighted works to train generative-AI systems is a topic for another day.
Key takeaways:
Copyright protects the elements of a work that are created by a human, but does not protect elements that were AI-generated (probably the key take-away from the Report) The is the “human authorship” requirement that the Copyight Office invoked in denying registration of Stephen Thaler’s AI-generated output. I wrote about that a couple of years ago in “AI Can Create But Is It Art?”
The Copyright Office believes existing law is adequate to deal with AI copyright issues; it does not believe any new legislation is needed
Using AI to assist in the creative process does not affect copyrightability
Prompts do not provide sufficient control over the output to be considered creative works.
Protection exists for the following, if they involve sufficient human creativity:
Selection, coordination, and arrangement of AI-generated output
Modification of AI-generated content
Human-created elements distinguishable from AI-generated elements.
Prompts
A key question for the Copyright Office was whether a highly detailed prompt could suffice as human creative expression. The Office says no; “[P]rompts alone do not provide sufficient human control to make users of an AI system the authors of the output. Prompts essentially function as instructions that convey unprotectable ideas. While highly detailed prompts could contain the user’s desired expressive elements, at present they do not control how the AI system processes them in generating the output.”
How much control does a human need over the output-generation process to be considered an author? The answer, apparently, is “So much control that the AI mechanism’s contribution was purely rote or mechanical. “The fact that identical prompts can generate multiple different outputs further indicates a lack of human control.”
Expressive prompts
If the prompt itself is sufficiently creative and original, the expression contained in the prompt may qualify for copyright protection. For example, if a user prompts an AI tool to change a story from first-person to third-person point of view, and includes the first-person version in the prompt, then copyright may be claimed in the story that was included in the prompt. The author could claim copyright in the story as a “human-generated element” distinguishable from anything AI thereafter did to it. The human-created work must be perceptible in the output.
Registration of hybrid works
The U.S. Copyright Office has now issued several registrations for works that contain a combination of both human creative expression and AI-generated output. Examples:
Irontic, LLC has a registered copyright in Senzia Opera, a sound recording with “music and singing voices by [sic] generated by artificial intelligence,” according to the copyright registration. That material is excluded from the claim. The registration, however, does provide protection for the story, lyrics, spoken words, and the selection, coordination, and arrangement of the sound recording.
Computer programs can be protected by copyright, but if any source code was generated by AI, it must be excluded from the claim. Thus, the Adobe GenStudio for Performance Marketing computer program is protected by copyright, but any source code in it that was AI-generated is not.
A record company received a copyright registration for human additions and modifications to AI-generated art.
As an example of a “selection, coordination and arrangement” copyright, there is the registration of a work called “A Collection of Objects Which Do Not Exist,” consisting of a collage of AI-generated images. “A Single Piece of American Cheese,” is another example of a registered copyright claim based on the selection, coordination, or arrangement of AI-generated elements.
China
A Chinese court has taken a contrary position, holding that an AI-generated image produced by Stable Diffusion is copyrightable because the prompts he chose reflected his aesthetic choices.
Internet Archives Postscript
In January, the Second Circuit Court of Appeals affirmed the decision in Hachette Book Group, Inc. v. Internet Archive. This came as no surprise. A couple of important things that bear repeating came out of this decision, though.
First, the Court of Appeals reaffirmed that fair use is an affirmative defense. As such, the defendant bears the burden of establishing the level of market harm the use has caused or may cause. While a copyright owner may reasonably be required to identify relevant markets, he/she/it is not required to present empirical data to support a claim of market harm. The defendant bears the burden of proof of a fair use defense, including proof pertinent to each of the four factors comprising the defense.
Confusion seems to have crept into some attorneys’ and judges’ analysis of the issue. This is probably because it is well known that the plaintiff bears the burden of proof of damages, which can also involve evidence of market harm. The question of damages, however, is separate and distinct from the “market harm” element of a fair use defense.
The second important point the Second Circuit made in Hatchette is that the “public benefit” balancing that Justice Breyer performed in Google LLC v. Oracle America, Inc. needs to focus on something more than just the short-term benefits to the public in getting free access to infringing copies of works. Otherwise, the “public benefit” in getting free copies of copyright-protected stuff would outweigh the rights of copyright owners every time. The long-term benefits of protecting the rights of authors must also be considered.
True, libraries and consumers may reap some short-term benefits from access to free digital books, but what are the long-term consequences? [Those consequences, i.e.,] depriv[ing] publishers and authors of the revenues due to them as compensation for their unique creations [outweigh any public benefit in having free access to copyrighted works.]
