June 1, 2024

U.S. Copyright Law Infringement

By Jill Hubbard Bowman

Generative AI Litigation

In the U.S., more than 14 cases have been filed involving AI technology in federal courts.[i]

  • Cases allege copyright infringement based on reproduction of input data used during the AI model training process.
  • Some plaintiffs also claim the generative AI models in the lawsuit and all output from the models are infringing derivative works of all of the training data, even if the model and output lack any similarity to the input data. A critical underlying allegation in these claims is that the generative AI model permanently incorporates the training data, which generative AI tool makers and many AI engineers adamantly deny.
  • Some plaintiffs also claim violation of the right of distribution and public display.
  • Additional non-copyright infringement claims include removal of copyright management information in violation of the DMCA, unfair competition, unjust enrichment, larceny of personal information, invasion of privacy, negligence, and trademark infringement.
  • The central copyright defense for using copyrighted content for AI training is fair use.

No cases have been brought against end users of AI generative tools, yet.

These AI copyright cases raise some of the most novel copyright questions in fifty years. Resolution in the courts will take many years. By then, AI will be everywhere and used by almost everyone-including artists,[ii]software programmers,[iii] news organizations,[iv] and lawyers.[v] Eventually, the government, AI developers, and content creators will likely negotiate ways to resolve the major disputes over use of copyrighted materials for AI training and mitigate the disruptions to the internet information ecosystem.

Below is a description of some of the basic U.S. legal concepts and unique AI infringement issues.

Copyright registration is required before bringing an infringement case in court.[vi]

The U.S. Supreme Court made a bright line rule that a U.S. copyright owner can’t bring a lawsuit in federal court until the copyrighted work is officially registered with the U.S. Copyright Office.[vii] Some who routinely register works before publication include publishing companies, music studios, and professional photographers. In contrast, ordinary people and many technology companies do not register their works, including software.[viii] Most copyrighted content on the internet is unregistered.

Violation of a copyright owner’s exclusive right may be direct infringement.

A copyright is not like a patent, which prevents “use” of an invention.[ix] The Ninth Circuit Court of Appeals stated “copyright protection may not serve as an invulnerable shield at the behest of any who invokes its powers.”[x] A copyright is only directly violated if someone does the following specific activities with a copyrighted work without authorization from the copyright owner or a defense: reproduction, creating a derivative work, distribution, public performance, or public display.[xi] Sometimes the term “copying” is used as short hand for a violation of any the exclusive rights.

Technology doesn’t really infringe. People do.

In the AI ecosystem, who may be a direct infringer along the AI development pipeline? Who is in control and performs the activity that may infringe a copyright? Is the use of copyrighted material in a dataset for AI training fair use? What are the implications for downstream users of AI technology from the acts of others upstream in AI development? What are the implications for AI developers from the acts of others who refine, modify and use AI technology they developed?

Although to date it is only generative AI makers who have been sued for copyright infringement, a central question will be whether only end users can be direct infringers for generating AI output alleged to infringe works by copyright holders. End users control the prompts into AI tools. An end user could even prompt some generative AI systems with entire copyrighted works by a third party. Who should be liable for the output if it is substantially similar to a copyrighted work? Who’s in control? A maker? A machine? Anyone?

A plaintiff must show copyright ownership and copying of protected elements.

To establish a claim of copyright infringement, a plaintiff must prove: 1) it owns a valid copyright, and 2) copying by the defendant of the protected expression in the plaintiff’s (registered) copyrighted work.[xii]

A plaintiff must show factual and legal copying of the copyrighted expression.

First, factual copying may be shown by 1) direct evidence of copying, or 2) indirect evidence of defendant’s access to the copyrighted work and probative similarities between the copyrighted work and the allegedly copied material.

But just because the fact of copying is established, the infringement analysis does not stop. The plaintiff also must show there was an actionable, legal violation.

Second, illegal copying is only shown if important, protected elements of the copyrightable work are copied. If only the non-expressive elements are copied, there is no actionable infringement.[xiii]

Evaluation of software infringement often requires an abstraction-filtration-comparison test to remove non-protectable elements from the analysis.

The copyright infringement analysis for any software is very difficult because of the intrinsic functionality of software and the large number of elements that are unprotectable. This makes the potential copyright protection for software programs thin. The distinction between an uncopyrightable process and a computer program performing that process is often elusive.

Courts struggle. A court may use an abstraction-filtration-comparison test to determine the proper scope of protection for computer programs as well as the Copyright Act’s exclusions in Section 102(b) and other copyright doctrines like the idea/expression merger doctrine and the scenes a faire doctrine.

The Tenth Circuit Court of Appeals explained that in computer cases, “in order to impose liability for copyright infringement, the court must find that the defendant copied protectable elements of the plaintiff's program and that those protectable elements comprise a substantial part of the plaintiff's program when it is considered as a whole.”[xiv]

Courts often make a distinction between literal elements of the software, the object and source code, and the other non-literal elements, like structure, sequence, and organization of the program. For the infringement analysis of non-literal elements of computer code, there is a common, three-part, infringement analysis process: 1) abstraction; 2) filtration; and 3) comparison.[xv]

Abstraction. First, in the abstraction step, the court breaks down the software program into its structural parts.

Filtration. Second, in the most important step, the unprotectable elements are “filtered” out and eliminated before the comparison. This is tricky because many elements in a computer program may be functional and unprotectable. Section § 102(b) of the Copyright Act is explicit: “In no case does copyright protection for an original work of authorship extend to any idea, procedure, process, system, method of operation, concept, principle, or discovery, regardless of the form in which it is described, explained, illustrated, or embodied in such work.”[xvi] The Copyright Office’s guidance on registration of computer programs clarifies that the computer program’s system design, algorithms, mathematical logic, formatting, command structures, and functions are not copyrightable.[xvii] The Copyright Office has also recently provided guidance about the requirement of human authorship. Only human-authored, creative expression is protected. There is no protection for traditional elements of authorship that are created by a mechanical process.[xviii] There is also no protection for human creative code that can only be written in one or a few ways, standard elements,[xix]common arrangements of information, [xx] or elements dictated by efficiency, external factors, or functional demands of the program.[xxi]

Comparison. The last step is to compare any remaining creative “golden nugget[s]” of human expression of non-literal elements in the software program to the allegedly infringing work. In some circumstances there is a higher standard than substantial similarity. Sometimes, a plaintiff must show virtual identicality between the works for legal copying to be proven and infringement found.[xxii] Similarly, “if technical or conceptual constraints limit the available ways to express an idea, even though there is more than one avenue of expression available, copyright law will abhor only a virtually identical copy of the original.”[xxiii]

Definition of “derivative work.”

“Derivative work” is a legal term of art specifically defined in the U.S. Copyright Act as “a work based upon one or more preexisting works, such as a translation, musical arrangement, dramatization, fictionalization, motion picture version, sound recording, art reproduction, abridgment, condensation, or any other form in which a work may be recast, transformed, or adapted. A work consisting of editorial revisions, annotations, elaborations, or other modifications which, as a whole, represent an original work of authorship, is a "derivative work.”[xxiv]

Proving infringement of the “derivative work” right may require showing the original elements of the pre-existing work were recast, transformed, or adapted.

For violation of the right to make a derivative work, a defendant must have recast, transformed, or adapted the elements of the pre-existing copyrighted work that make it original. This analysis seems to be at a higher level of abstraction than the analysis of for the right of reproduction. [xxv]

If the defendant only copied and did not recast or transform the original work, then the analysis should be under the section related to the right of reproduction not derivation.

“The legislative history of the Copyright Act of 1976 indicates that Congress intended that for a violation of the right to prepare derivative works to occur "the infringing work must incorporate a portion of the copyrighted work in some form."[xxvi] The Nimmer treatise explains that to qualify as a derivative work, the work must include the copyrightable elements. It states, a “work will be considered a derivative work only if it would be considered an infringing work if the material which it has derived from a preexisting work had been taken without the consent of a copyright proprietor of such preexisting work.”[xxvii]The copyrightable elements do not include ideas, information, methods, or math. If only non-expressive, uncopyrightable elements of a copyrighted work are recast, transformed, or adapted, the new material isn’t legally an infringing “derivative work.”

Infringing Derivative Works

Critical questions in the AI debate over derivative works:

  • Does an infringing derivative work differ from a copyrightable derivative work?
  • Does an infringing “derivative work” have to be authored by a human and be original?

U.S. courts and legal scholars don’t agree on the answers.

Some legal commentators and courts (Ninth Circuit) make a distinction between an infringing derivative work and a copyrightable derivative work, which must meet all requirements for copyrightability. Others including the Nimmer copyright treatise and the Seventh and Eleventh Circuit Court of Appeals don’t. They require originality in the infringing derivative work.[xxviii]

In Lee v. A.R.T. Co., the Seventh Circuit Court of appeals noted the split in the courts on the requirement of originality for an infringing derivative work. It described the lower court’s analysis for its finding that bonding an art card to a tile was not a “derivative work”:

“Pointing to the word "original" in the second sentence of the statutory definition, the district judge held that "originality" is essential to a derivative work. This understanding has the support of both cases and respected commentators. E.g., L. Batlin Son, Inc. v. Snyder, 536 F.2d 486 (2d Cir. 1976); Melville B. Nimmer David Nimmer, 1 Nimmer on Copyrights sec. 3.03 (1997).”

The Seventh Circuit noted the conflict with the Ninth Circuit on the requirement of originality:

“Pointing to the fact that the first sentence in the statutory definition omits any reference to originality, Lee [the plaintiff] insists that a work may be derivative despite the mechanical nature of the transformation. This view, too, has the support of both cases and respected commentators. E.g., Lone Ranger Television, Inc. v. Program Radio Corp., 740 F.2d 718, 722 (9th Cir. 1984); Paul Goldstein, Copyright: Principles, Law and Practice sec. 5.3.1 (2d ed. 1996) (suggesting that a transformation is covered by sec. 106(2) whenever it creates a "new work for a different market").  . . .

If Lee (and the ninth circuit) are right about what counts as a derivative work, then the United States has established through the back door an extraordinarily broad version of authors' moral rights, under which artists may block any modification of their works of which they disapprove. No European version of droit moral goes this far. Until recently it was accepted wisdom that the United States did not enforce any claim of moral rights; even bowdlerization of a work was permitted unless the modifications produced a new work so different that it infringed the exclusive right under sec. 106(2).”

The Seventh Circuit found it “jarring” that the plaintiff’s attorneys in oral argument agreed that merely handwriting on the copyrighted art card or using it as a coaster for a drink would be a derivative work. The court was not impressed that the plaintiff’s attorneys reassured the court that artists wouldn’t sue for making those types of “derivative works.”

Regardless of the lack of clarity in the statutory definition, the Seventh Circuit found that the plaintiff’s argument that merely adhering a purchased print to a tile failed to meet any part of the statutory definition of “derivative work.” The work was not recast, adapted, or transformed.[xxix] In a later case, the Seventh Circuit referred to the originality requirement articulated in Lee as essential in the infringement analysis of derivative work cases.[xxx]

In a similar case of fact with a highly criticized opinion, which ignores the first sale doctrine, the Ninth Circuit Court of Appeals in Mirage Editions Inc. v. Albuquerque A.R.T. Co. ruled that gluing a print from an art book that the defendant had purchased was a “derivative work” because it made “a new version of the work.”[xxxi]No originality existed.

The Eleventh Circuit Court of Appeals rejected the reasoning of the Ninth Circuit in Mirage Editions and instead aligned with the Seventh Circuit in Lee. It found that the alleged derivative work as a whole lacked originality and rejected the copyright infringement claim.[xxxii]

 Law Professor Daniel J. Gervais in his article, AI Derivatives: The Application to the Derivative Work Right to Literary and Artistic Productions of AI Machines, suggests that the “key take away” from these derivative work cases and his analysis is that “under the Ninth Circuit test, a machine could infringe 17 U.S.C. §106(2) but not under the Seventh (or Eleventh’s) Circuit’s.”[xxxiii]

 Under the view that originality is required for an infringing “derivative work,” if AI technology (like AI generated output, AI model weights autonomously created by deep learning algorithms, and machine-generated data) doesn’t include original expression authored by a human, it can’t infringe the derivative work right.

 This seems like an issue for the U.S. Supreme Court to resolve.

All output from AI models can’t be infringing derivative works.

Human expression doesn’t come out of thin air. What we express draws on what we have learned over our lifetimes. Expressive language is based on language we have encountered in many different forms and forums. We have encountered hundreds of thousands of words. Not everything we write is a “derivative work.” Indeed, independent creation is a defense to copyright infringement.

Similarly, some large language models (LLMs) have learned from billions of words–the content of the internet–to create a knowledge model. The models have learned the unprotectable concepts and structure of language, which are embodied in the weights (its brain). The model uses statistics to predict words to write in response to a prompt. The LLM does not contain the training dataset or processed tokens. The trained model is not a translation, compression, or abridgement of the expressive content in training data and shouldn’t be considered an infringing derivative work.

Assuming human originality is not required for output to be considered an infringing derivative work, whether output could potentially infringe a pre-existing work used for training as a derivative work, would depend on whether it recast, transformed, or adapted the original expressive choices of the original training work and is substantially similar. Whether there is any issue with possible copyright infringement for output from an AI tool in large part depends on whether the AI tool was trained or refined only with licensed data. If unlicensed copyrighted data was used in training or a user inputs prompts consisting of a third-party’s copyrighted work into a generative AI tool, then potential copyright infringement issues related to output may arise depending on the type of model and its output.

Regardless of the status of the training data input, most output from an AI generative tool, however, is new content. If the output isn’t substantially similar to any copyrighted content, there isn’t a solid argument for copyright infringement of any kind for the output. How an AI model was trained should be a completely distinct infringement issue from output, regardless of the fruit-of-the-poison-tree/derivative work type arguments some plaintiffs are making against generative AI output.

AI generative tools are not search engines. For example, large language models (LLMs) like ChatGPT don’t contain training datasets. Once they are trained, they don’t have access to the training datasets. After a prompt, the LLMs do not go back and search an internal database for an answer. Rather, the trained knowledge model of the AI system uses math and statistics to guess what the output, like the next word in a sentence, should be. Sometimes, the LLMs are really good guessers.

Prompting a LLM to generate output may be alleged to be copyright infringement.

In their response to the Copyright Office’s notice of inquiry, Microsoft and GitHub admitted that copyright holders have legitimate concerns over substantially similar output from generative AI systems, which they conceded might be infringing. They are adamant, however, that training an AI model on unprotectable concepts is not the same as publishing a substantially similar work.

Microsoft and OpenAI have conceded that, despite their efforts to mitigate the risk of substantially similar outputs, sometimes their models may generate output similar to training data “where the training data contains the same words appearing in the same order in multiple places and contexts.” It’s a type of “bug” in the AI system. [xxxiv]  When very specifically prompted by a user to elicit particularly over-sampled text, the output may be part of the original text.

If an end user of a generative AI tool prompts it to output verbatim or substantially similar text to a known copyrighted work, the end user might be alleged to be a direct copyright infringer. For example, if the user prompts an AI tool by saying, “Show me the full text of X article from the following link (and provides the website link to the article”), a plaintiff may argue that the end user is a direct infringer if the tool has inadvertently memorized the entire article and provides the copyrighted text verbatim. Just like a person can’t use Microsoft Word to copy and paste an entire copyrighted article from a link without possibly running afoul of the exclusive right of reproduction, an end user likely can’t trick and use an AI tool to do the same thing. A fair use defense would still be available if an infringement claim was brought by a copyright holder. Whether copying the output was fair use would be a fact specific inquiry under the four-factor fair use test.

 Software developers have less cause for concern. GitHub Copilot is a generative AI tool that turns natural language prompts into software code in many different software languages. An answer in a FAQ on the Copilot website, says that their research indicates the suggested output code “matches about 1% of public code.”[xxxv] Some of this matching code is likely uncopyrightable code because that software can only be written in one or a few ways or is very commonly used in the industry. For some widely used code sequences it would be difficult to even identify an original author who might own the copyright since such common code is replicated and occurs frequently on public repositories like GitHub. Copilot has a filter setting to prevent public repository matching code from being outputted, eliminating most copyright infringement arguments. GitHub also added a feature for developers who allow matching code suggestions by the tool to identify the matching public code on GitHub. This allows a developer to view the original license terms, helping qualm the concerns of users over copy-left, open-source licensed code.[xxxvi]

Copyright indemnification from AI tool providers is sometimes available.

Given the current AI copyright circus and the consumer liability fears, some AI tool providers like Microsoft, GitHub, and Adobe are providing IP indemnity for their customers if they are sued for copyright infringement for copying and distributing the output of their AI tools. The indemnity terms, however, are narrow and the customer must have adhered to their Terms of Use when using the tool. AI tool makers are playing the odds. It’s unlikely an end user with shallow pockets and undetectable use will be sued for copyright infringement for reproducing output. Litigation is too expensive and hard. And proving substantial similarity for most output if the AI tool is used responsibly will be difficult if not impossible in court.

Contributory infringement is inducing direct infringement.

An entity connected to direct infringement may also be liable for copyright infringement if they knew of the infringing activity and materially contributed to or caused the infringement.[xxxvii] In the context of the long complex process for developing AI technology, which involves numerous actors and technology providers, this is a concern for companies enabling the AI developer ecosystem and those providing services or products from third-party AI technology.

Vicarious infringement is based on controlling and profiting from infringement.

Vicarious infringement involves the right and ability to control the infringing activity, and a direct financial interest in the infringement.[xxxviii] This is how plaintiffs are trying to snare related corporate entities or companies working together to create new AI technology.

Inducing infringement is based on intending to contribute to the infringement.

Inducing infringement involves making a material contribution to the infringing activity with the intent to cause infringement.[xxxix] This is a more difficult claim since it requires the intent to infringe.

Damages may be actual damages/infringer’s profits, or statutory damages.

In a copyright infringement action, the copyright owner may recover actual damages they suffer as a result of the infringement (lost profits) or any additional profits of the infringer, or in some limited circumstances, statutory damages.[xl]

Early registration is required to obtain statutory damages in federal court.

If a work is not registered before infringement or within 3 months of publication, statutory damages are not available in federal court. This is significant because without statutory damages the plaintiff can only recover actual damages or profits attributable to the infringement of its work, which is often difficult to prove for an individual work. Most works on the internet are not registered and owners could not get statutory damages.

Statutory damages vary between $200 to $150,000 per work based on intent.[xli]

Statutory damages are typically between $750 and $30,000 per work infringed (not by the number of times a work was infringed.) Plaintiff may also recover attorneys’ fees and costs.

Innocent infringement:  If the court finds the infringer was not aware or had no reason to believe the activity was infringement, the court may lower the statutory damages to a minimum of $200 per work.

Willful infringement:  If a court finds the infringer knew or recklessly disregarded the possibility the activity was infringing, then statutory damages may be up to $150,000 per work.

Fair Use Doctrine

The U.S. Copyright Act in 17 U.S.C. § 107 explains the framework for a legal doctrine that promotes freedom of expression by permitting the unlicensed use of copyrighted works in some circumstances. Congress was explicit that using a copyrighted work for the purpose of teaching, research, scholarship, criticism, comments, and news reporting was not copyright infringement under the statute.

Historically, U.S. courts have applied the fair use doctrine when confronted with transformative technology. The fair use doctrine is open ended and flexible, but it is highly factual and case specific. Courts have allowed reproductions of libraries of books,[xlii] thumbnail images on the internet,[xliii] and APIs for a programming language. It can take as long as a decade for a case to wind its way through the U.S. judicial system for resolution of a high stakes copyright case with a fair use defense. And the courts and juries along the way may have very different analysis of the fair use factors and come to the opposite conclusion as to whether a use of copyrighted work is fair.[xliv] Only the U.S. Supreme Court can decide definitively whether a use of a copyrighted work is fair under the law.

Although informative to a legal analysis of fair use and to a plaintiff’s lawyer weighing chances for success in court in a contingency-fee based, infringement action, no broadly applicable rule for all AI training will result from the pending generative AI copyright infringement cases.

Fair Use Factors

Overall, courts try to balance the interests of the plaintiffs with the interests of the public and copyright policy. With highly disruptive AI technology that will radically transform the world of work and creation while launching new markets through AI development, balancing interests will be difficult.

The U.S. Copyright Act outlines a four-factor test for determining whether use of a copyrighted work is fair:[xlv]

  1. The purpose and character of the use
  2. The nature of the copyrighted work
  3. The amount and significance of the copied portion
  4. The market effect on the copyrighted work

A finding of fair use should be easier for the training of non-generative AI models where the output may simply be the identification of an object like a hard hat on a construction worker or a weld defect on a car part in a factory. The purpose of the use of copyrighted works, which may be in a training dataset, in AI training is to learn from non-protected elements of the work. The model is trying to learn the information in the work so it can do things like identify objects accurately. It doesn’t care about any creative expression in something like a photograph of a hard hat or image of a weld defect. (Factor 1) There is a strong argument that AI models are highly innovative technology and a transformative, non-expressive use that will promote science and the arts for the public good. Use of a broad data set makes an AI model more accurate, safe, and efficient. (Factor 2) Although entire works may be copied for training, there is not even a question for most neural network models that the model itself does not contain the training data. The non-expressive output of an identification, recommendation, or a decision will not harm the market for the expressive elements of the copyrighted work. (Factor 4). Therefore, depending on the specific facts and types of actors, a defendant in an infringement case related to non-generative AI models may have strong arguments under the fair use test and may prevail.

Even for generative AI models that can produce expressive content, fair use for AI training on copyrighted content is a defense that is likely to prevail in many cases. Mark Lemley, Professor and Director of Stanford’s Law, Science, and Technology, in his law review article, Fair Learning, makes a persuasive case there should “generally” be a clear exception in copyright law to allow AI training with copyrighted works. First, massive datasets are needed to make very large AI models accurate, reliable, unbiased, and fair. It would be virtually impossible to license all the content needed for robust AI training from millions of copyright holders for billions of works. It’s impractical. There simply are not enough datasets available for meaningful licensing. Allowing copyright claims would not mean all copyright holders would be paid. Rather, allowing copyright claims would stunt AI development of small companies and limit advancement to only the biggest, richest companies who could pay lawyers for a decade of expensive litigation to prove fair use or settle with those who can afford to sue. Second, most machine learning systems, including large language models, are not using copyrighted works for creative expression, or sharing that expression with the public. Rather, copyrighted works are used for a non-expressive, transformative purpose-learning. In AI training, works are used for teaching the AI model about the unprotected information and concepts in the works about language and its structure. And the intent is to create new content.

It’s unclear whether U.S. courts, the U.S. Copyright Office, or Congress will first answer the question of whether organizations can freely use copyrighted works without permission for AI training or what the analysis should be to examine allegedly infringing AI output. What is clear is that copyright holders do not have the right to prevent all use of a work they author. The public interest in use of information for learning needs to be protected from copyright lock in of unprotected elements of a work.

It is also clear that allowing development of powerful AI will benefit society broadly by tackling big problems in healthcare, education, power, and climate change. Although AI is going to cause a major paradigm disruption for content creators, AI tools will also help creators be more creative and efficient. Other laws should be passed to address concerns related to potential abuses of AI output in mimicking people’s likeness, style, and images. If the U.S. does not allow AI training, it will lag behind other countries with more expansive copyright policies putting U.S. national security and economic growth at a disadvantage. Given the speed of world development of powerful, earth-changing AI technology, which has been likened to a nuclear bomb by Microsoft AI’s CEO,[xlvi] the U.S. does not want to lose the AI race.

Jill Hubbard Bowman is an intellectual property attorney with over 25 years of legal experience representing software companies. She has litigated software copyright, patent, and trade secret cases across the U.S. and has represented plaintiffs and defendants. For many years, Jill has advised companies on  AI technology.  Jill has also worked with corporate strategy and execution teams on ethical design and deployment of AI technology. 


[i] See the valuable AI litigation tracker by law firm Baker Hostetler at https://www.bakerlaw.com/services/artificial-intelligence-ai/case-tracker-artificial-intelligence-copyrights-and-class-actions/

[ii] See Adobe Firefly AI image generator. If you can’t find an appropriate image for your work on Adobe Stock, use generative AI to make your own, like the first image in this article and many other AI images on Adobe Stock.

[iii] GitHub announces more than a million paid subscribers and 37,000 organizations using its Copilot AI software tool. https://www.linkedin.com/posts/ashtom_microsoft-has-over-a-million-paying-github-activity-7123051140115562497-7Hbq/

[iv] Washington Post plans to aggressively implement AI in its business. https://www.thewrap.com/washington-post-77-million-loss-ai/

[v] Stokel-Walker, Chris,“Generative AI is Coming for Lawyers,” Wired, Feb. 21, 2023,  https://www.wired.com/story/chatgpt-generative-ai-is-coming-for-the-lawyers/

[vi] The Copyright Claims Board is for small copyright disputes. Total damages cannot exceed $30,000 per case. Early registration is not required for statutory damages but limits the claim amount up to $7,500 per work and up to $15,000 per case.

[vii] Reed Elsevier, Inc. v. Muchnick, 599 US 154 (2010).

[viii] In part, the copyright registration process for software is awkward and time consuming. Registration requires a deposit of the code (trade secrets may be blacked out). Usually, registration only covers one code version, which is cumbersome when code is updated quarterly. You must prove a valid copyright registration for the work you allege to infringe in court. The registration can’t be to a different version of the software, or the case will be dismissed.

[ix] In the U.S., a patent right gives the owner the right to exclude others from making, using, selling, offering to sell, or importing the patented invention, which is described in a patent claim.

[x] Apple Computer, Inc. v. Microsoft Corp., 35 F.3d 1435, 1446 (9th Cir. 1994).

[xi]  17 U.S.C. § 501 https://www.copyright.gov/title17/92chap5.html

[xii] Feist Publications v. Rural Telephone Service Co., 499 U.S. 340, 361, 111 S.Ct. 1282, 1296, 113 L.Ed.2d 358 (1991) (a foundational U.S. Supreme Court copyright opinion explaining infringement analysis).

[xiii] MiTek Holdings, Inc. v. Arce Eng’g Co., 89 F.3d 1548, 1554 (11th Cir. 1996) (finding menu and submenu command tree structures to be an uncopyrightable processes).

[xiv] Gates Rubber Co. v. Bando Chem. Indus., Ltd., 9 F.3d 823, 832 (10th Cir. 1993).

[xv] Comput. Assocs. Int’l, Inc. v. Altai, Inc., 775 F. Supp. 544 (E.D.N.Y. 1991), aff’d, 982 F.2d 693 (2d Cir. 1992) (the leading case on copyright analysis of software filtering out elements dictated by efficiency, constrained by external factors, and commonplace programming techniques). See an overview of all tests related to copyright infringement analysis for software by Samuelson, Pamela, Berkeley Technology Law Journal, “Functionality and Expression in Computer Programs: Refining the Tests for Copyright Infringement.” https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2909152

[xvi] 17 U.S.C. § 102(b); Gates Rubber Co., 9 F.3d at 842–843.

[xvii] U.S. Copyright Office, Guidance on Copyright Registration of Computer Programs, https://www.copyright.gov/circs/circ61.pdf; Sega Enterp. Ltd. v. Accolade, Inc., 977 F.2d 1510, 1522, 1526 (9th Cir. 1992); Compendium of U.S. Copyright Practices (3d Ed. 2015) (hereinafter “Compendium”), § 721.7 (“the Office will not register the functional aspects of a computer program, such as the program’s algorithm, formatting, functions, logics, system design, and the like.”) See also id. at § 721.9(J) (indicating that registration will not be accepted if the program author claims copyright in an algorithm, computation, data, formatting, formulas, interfaces, language, layout, logic, menu screens, models, organization, protocols, and system design, among other categories of unprotectable elements).

[xviii]U.S. Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence https://copyright.gov/ai/ai_policy_guidance.pdf

[xix] Corwin v. Walt Disney Co., 475 F.3d 1239, 1251 (11th Cir. 2007); see also Beal v. Paramount Pictures Corp., 20 F.3d 454, 459 (11th Cir. 1994) (describing scènes à faire as "stock scenes that naturally flow from a common theme").

[xx] Feist Publications, at 361.

[xxi] Altai, 982 F.2d at 707–09, 714-715.

[xxii] Apple Computer, 35 F.3d at 1446.

[xxiii] Id.

[xxiv] 17 U.S.C. § 101 (emphasis added. Definition of “derivative work”). https://www.govinfo.gov/content/pkg/USCODE-2011-title17/pdf/USCODE-2011-title17-chap1-sec101.pdf.

[xxv] Mirage Editions Inc. v. Albuquerque A.R.T. Co., 856 F.2d 1341 (9th Cir. 1988) (finding that gluing an image from a purchased book to a tile was an infringing derivative work because it “made another version” of the copyrighted art work) https://law.resource.org/pub/us/case/reporter/F2/856/856.F2d.1341.87-6465.html. See also 17 U.S.C. §§ 101, 106.

[xxvi] Mirage at p. 18, citing 1976 U.S. Code Cong. & Admin. News 5659, 5675. (emphasis added).

[xxvii] Melville B. Nimmer & David Nimmer, 1 Nimmer on Copyrights § 3.01 (1997) cited in Litchfield v. Spielberg, 736 F.2d 1352, 1357 (9th Cir.1984), cert. denied, 470 U.S. 1052, 105 S.Ct. 1753, 84 L.Ed.2d 817 (1985); United States v. Taxe, 540 F.2d 961, 965 n. 2 (9th Cir.1976), cert. denied, 429 U.S. 1040, 97 S.Ct. 737, 50 L.Ed.2d 751 (1977).

[xxviii] Daniel Gervais, “AI Derivatives: The Application to the Derivative Work Right to Literary and Artistic Productions of AI Machines,” 52 Seton Hall L. Rev. 1111 (2022) (hereinafter “AI Derivatives’)

[xxix] Lee, 125 F.3d at 580.

[xxx] Picket v. Prince, 207 F.3d 402 (7th Cir. 2000).

[xxxi] Mirage Editions Inc. v. Albuquerque A.R.T. Co., 856 F.2d 1341 (9th Cir. 1988) (finding that gluing an image from a purchased book to a tile was an infringing derivative work because it “made another version” of the copyrighted art work) https://law.resource.org/pub/us/case/reporter/F2/856/856.F2d.1341.87-6465.html. See also 17 U.S.C. §§ 101, 106, AI Derivatives, at p. 10.

[xxxii] Peter Letterese & Associates, Inc. v. World Inst. Of Scientology Enters., 533 F.3d 1287, 1299 (11th Cir. 2008).

[xxxiii] AI Derivatives, at p. 19.

[xxxiv] https://www.regulations.gov/comment/COLC-2023-0006-8906

[xxxv] https://resources.github.com/learn/pathways/copilot/essentials/how-github-copilot-handles-data/

[xxxvi] https://www.regulations.gov/comment/COLC-2023-0006-8750

[xxxvii] A & M Recs., Inc. v. Napster, Inc., 239 F.3d 1004 (9th Cir. 2001).

[xxxviii] Fonovisa, Inc. v. Cherry Auction, Inc., 76 F.3d 259, 263 (9th Cir. 1996).

[xxxix] Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd., 545 U.S. 913 (2005).

[xl] 17 U.S.C. § 504 a. https://www.copyright.gov/title17/92chap5.html

[xli] 17 U.S.C. § 504 c.

[xlii] Authors Guild v. Google, Inc., 804 F.3d 202, 207 (2d Cir. 2015).

[xliii] Kelly v. Arriba Soft Corp., 336 F.3d 811 (9th Cir. 2003).

[xliv]See the ping-ponging nature of the Java API cases and final result in Google LLC v. Oracle America Inc., 141 S. Ct. 1163 (2021) https://www.supremecourt.gov/opinions/20pdf/18-956_d18f.pdf

[xlv] 17 U.S.C. § 107 (fair use test)

[xlvi] Mustafa, Suleyman, The Coming Wave – Technology, Power, and the 21st Century’s Greatest Dilemma (Referencing restricting deployment of very powerful AI: “If everyone in the world can play with nuclear bombs, at some stage you have nuclear war.”)


This AI Law Maze Map blog is for education only. It is not intended as legal advice.

By using this website and information, you acknowledge and agree that no attorney-client relationship is created or implied.

Sign up for our newsletter

© 2023 Jill Hubbard Bowman. All rights reserved.