Tag: Fair use

AI Training and Copyright: It’s a Global Issue—and There is a Global Solution

Infographic titled 'AI Training & Copyright: How Jurisdictions Approach AI Training and Copyrighted Works' detailing various jurisdictions including the USA, European Union, Japan, South Korea, China, United Kingdom, Canada, and Australia. It summarizes their legal frameworks, approaches to AI training with copyrighted works, and key points regarding copyright considerations.

Image: Used with permission (Maryna Hryhorieva)

This striking graphic, created by IP lawyer Maryna Hryhorieva, is a great summation of the current state of play regarding the still very fluid issue of AI training and copyright. It compares how eight different regimes, the US, EU, Japan, South Korea, China, the UK, Canada and Australia, are addressing this thorny issue, and highlights the different frameworks used to determine what is legal and what’s not. Each is different with its own peculiarities. Navigating through them is not easy. What may apply in one jurisdiction will not apply in another. A use that is acceptable in Country A may result in an unfavourable court ruling in Country B. An AI company can decide to base its content use practices on the most AI-friendly regime in which it operates, staying onside in that jurisdiction, but then can still be found to be infringing elsewhere. Cross-border jeopardy is almost certain where an AI company wants to scoop up local cultural content, pitting it against determined rightsholders who will, among other things, invoke cultural sovereignty.

The essence of the issue goes back to the way in which international copyright law is structured. Each country sets its own rules, regulations and interpretations within a broad commonly agreed framework. That framework, based on the Berne Convention, results in common basic standards centred on some fundamental principles, incarnated in the famous “three step test”.[i] The result is a generally consistent pattern of protection for rightsholders but with some significant differences by jurisdiction, like duration of copyright protection for example. The same inconsistencies apply to data mining (or what I would call unauthorized content reproduction) for AI training. So, what is a poor little ol’ trillion-dollar AI developer to do? Guess what? There is a universally applicable solution. Can you guess what it is? I reveal all in the final paragraph.

First, let’s briefly summarize the chart.

The US approach is based squarely on litigation, letting the courts decide the parameters of fair use regarding the unauthorized taking of protected content for AI training. This case-by-case approach will provide case-by-case answers, depending on the circumstances, the judge in question and even the circuit within the US federal justice system in which the case is heard. A legal approach inevitably leads to risk and uncertainty. That is not to say that in the end, the US will not resort to legislation. Thee US Justice Department recently submitted a “Statement of Interest” in the OpenAI v New York Times case, stating that a ruling in favour of the Times and other news industry plaintiffs would “threaten US national security, give a competitive advantage to foreign adversaries”, and thwart “creative and scientific progress while hindering American prosperity”. The Statement of Interest tries to put its thumb on the scale of justice, weighing in in support of the OpenAI argument that its use of NYT content was “transformational”, and thus constituted fair use. The Court is under no obligation to accept this interpretation; its duty is to interpret the law as it stands. But if the DOJ statement represents the Trump Administration’s interpretation of the law, and that interpretation is not upheld by the courts, that potentially opens to door to legislative change, assuming that Congress can be convinced to act in the name of big tech. Either way, there is continued uncertainty.

The EU approach is based on a narrow text and data mining exception embedded in the Copyright Directive. Resort to the exception is limited in two ways; if rightsholders cannot opt-out, the exception is limited to scientific organizations and cultural heritage institutions for research purposes only. There is also a general exception that allows for commercial use but in this case, rightsholders can reserve their rights by opting out.

Japan is considered by many to have a very AI-friendly regime in terms of allowing copying for training purposes, but it is not as wide as many have claimed, as I pointed out in this blog posting a couple of years ago. South Korea is struggling with the issue, being whipsawed between AI developers promising the moon and its vibrant cultural industry. China likewise is still trying to determine which way to go. It does not have a text and data mining (TDM) in its law, nor does it apply a fair use doctrine. However, it has announced plans to deal with AI training rules through its copyright development plan covering the period 2026-30.

The UK has had several false starts. A TDM exception limited to non-commercial purposes only has been on the books for several years, but the Labour government tried to expand this by advocating for a regime that would allow AI developers to freely use copyrighted content for algorithm training unless rightsholders opted out. After a massive outcry and pressure from British cultural industries and prominent performers, the government retreated. Now Britain has gone back to the drawing board.

In the case of Canada, everything is up in the air. Canada has a fair dealing regime with no TDM exception. It has held several public consultations and is developing an AI strategy but has been very unclear about where the government comes down on the issue of freely using copyrighted material for AI training. Meanwhile several Canadian media outlets are suing OpenAI for copyright infringement. OpenAI lost its argument that Canadian courts had no jurisdiction in this case. Now the Canadian music collective SOCAN is suing AI music company Suno. While AI companies are arguing for a wide TDM exception in Canadian law, rightsholders are seeking to retain the current law and use it to protect their rights.

Australia on the other hand has spoken out strongly in defence of its cultural sector and has declared that it will not legislate a TDM exception. Australian Prime Minister Albanese has taken that a step further declaring that, “An artist’s creative endeavour is their work and their property. No company should use Australian books, music, art or news to build or train AI without the artist’s control. That includes the artist’s control of the price and value of their work.” What will happen next in Australia is unclear although some tech platforms continue to try to push various compulsory licensing regimes.

As you can see, the rules are diffuse and in flux. Frankly, it is all over the map. One thing seems to be clear. The push by the tech industry for widespread TDM exceptions seems to be stalling. The litigation route so favoured in the US is yielding uncertain results. Legal actions in fair dealing countries like Canada, the UK and Australia may bring varying results with complex rules governing applicability of judgements. All this leads to greater uncertainty, and costs. However, as mentioned above, there is one universal remedy, one that transcends all the jurisdictional issues and conflicting interpretations. It is called voluntary licensing.

Voluntary licensing between rightsholders and AI developers eliminates the jurisdictional problems, assuming the agreements are structured to provide wide coverage and indemnity. With a voluntary licence, it doesn’t matter if there is or isn’t a text and data mining exception, or if the use is for commercial purposes, or if the use is sufficiently transformational. It doesn’t matter whether the AI developer is truly “enjoying” (accessing the essence of the content), as in the case of Japan. Some jurisdictions have toyed with the idea of a compulsory licence, as in India, but not only has this been opposed by both rightsholders and the AI industry, it is jurisdictionally bound. Voluntary licences deal with this issue and, if properly structured, remove the hazard of conflicting jurisdictional rulings.

While it is not always easy to reach such deals, as the OpenAI v New York Times case illustrates, it is surely better and less costly than the alternative of endless litigation. This is why it is becoming the preferred solution. More and more licensing agreements are being negotiated across the full range of copyrighted content, including audio-visual, music and publishing industries for AI training and use. Voluntary licensing is a global solution–but it requires AI companies to recognize the value of the content they want for training, and to accept the right of creators to control the use of their content. If it requires a willing buyer, it also requires a willing seller and while not all rightsholders will be willing to sign a licensing agreement, the holdouts will be the exceptions if fair deals are offered. It is a far better solution for both sides, AI developers and rightsholders, than endless litigation or constant lobbying for (or pushing back against) legislative change. It’s a global solution to a global problem.

© Hugh Stephens, 2026. All Rights Reserved.

[i] Any exception must satisfy three requirements: (1) Limited to certain special cases:. i.e. the exception must be narrowly defined and clearly circumscribed, rather than acting as a broad or general exemption; (2) No conflict with normal exploitation: i.e.the use must not interfere with the ways the copyright owner routinely makes money or manages the market for their work; (3) No unreasonable prejudice to legitimate interests of rightsholders, i.e. the use must not cause unfair or excessive economic or legal harm to the author or right-holder

Pete Hegseth and Franklin the Turtle: There is a Parody, Just Not the One You Think

Cover of a fictional book titled 'Franklin Targets Narco Terrorists', featuring a cartoon turtle in military attire, using a weapon from a helicopter while targeting boats.

Image: X/@PeteHegseth. Not copyright protected (produced with AI)

Pete Hegseth, US Secretary of War (or is it still Defense?) has demonstrated a deft capacity for stepping firmly into one cowpoop after another. There was the fiasco of discussing imminent strikes against Houthi rebels in a Signal chat group that included his wife, brother and personal lawyer, along with a few others such as journalists not authorized to have access to such high security information. Now there is the issue of whether he issued a second kill order for two survivors of a drug boat that had just been targeted by the US military. As serious as this may be, his subsequent ridiculing of the incident by co-opting friendly old Franklin the turtle to make light of his actions demonstrates not only appalling judgement, but also a cavalier attitude to and ignorance of trademark and copyright law that you would have thought that a senior government official of cabinet rank might just take into consideration. But this was not a communique issued by the Defense Department, it was a personal tweet on X. If anything, it underlines the risks of senior government officials (from the President on down) using social media accounts to comment on government policy.

The fact that Hegseth saw nothing wrong with appropriating the trademarked and copyrighted character of Franklin for his own ends while manipulating it with AI, in the process potentially damaging the brand image of a well known and well-loved children’s literary and cartoon icon, shows the level of his ignorance, hubris and disrespect. The fact that he chose a Canadian children’s character may not have been an accident. Would he have taken on Disney by putting the bazooka into the hands of Snow White or Warner Bros by having Bugs Bunny finish off the alleged drug traffickers. Probably not, although it is apparent that his ignorance of the law runs pretty deep.

Franklin’s publisher Kids Can Press, based in Toronto, issued the following statement:

“Franklin the Turtle is a beloved Canadian icon who has inspired generations of children and stands for kindness, empathy, and inclusivity…We strongly condemn any denigrating, violent, or unauthorized use of Franklin’s name or image, which directly contradicts these values.”

Will they sue? Likely not, since Corus, which exploits the character created and copyrighted by author Paulette Bourgeois and illustrator Brenda Clark, is in deep financial trouble, but they probably could. If they did, Hegseth (who would be the defendant since he posted the image on his personal account) would no doubt claim the US fair use defence of parody. But is this a parody?

To qualify as a parody under fair use in the US, (Section 107 of the US Copyright Act), it must be used for criticism and comment (of Franklin). As outlined in detail in this interpretive post from the Washington, DC based Copyright Alliance, “both parody and satire use humor as a tool to convey a message, but each serves a different purpose. Parody imitates the style of a particular creator with deliberate exaggerations for comedic effect. Satire uses humor to comment on the world-at-large, particularly in the context of politics. While both parody and satire incorporate criticism and commentary, only parody may be considered fair use.”

In other words, Hegseth would have to rely on parody as a fair use defence for his unauthorized use of the copyrighted images and name of Franklin. But is he parodying Franklin the Turtle, the ultimate “nice guy”? Maybe that is his intent, “No more Mr. Nice Guy”. The character is also trademarked, leaving open the possibility of a suit on the grounds of trademark impairment. After all, if Franklin is at heart is a remorseless killer, do you want your kid reading the books or watching the animated shows?

There have been famous cases where US courts ruled in favour of a parody defence, notably what is referred to as the Campbell v Rose-Acuff Music case. As explained by the US Copyright Office, the plaintiff, a music publisher and co-owner of the Roy Orbison 1964 rock ballad “Oh, Pretty Woman”, accused the defendants, hip-hop group 2 Live Crew, of infringing Orbison’s song by releasing a 1989 parody version titled “Pretty Woman.” On appeal the US Supreme Court overturned a lower court ruling that had denied fair use because of the extensive amount of material copied. The USCO stated it is essential for someone doing a parody to be able to quote from existing material and use some of the elements of a prior work to create a new one that comments on the original. On the other hand a full-length knock-off/parody of Dr. Suess, “Oh the Places You’ll Boldly Go”, setting Suess characters into a Star Trek world, was ruled an infringement, again on appeal.

So, was Pete Hegseth’s unauthorized use of Franklin to promote his version of the war on drugs on his personal social media account a parody? We will probably never know.

Franklin’s Wikifandom site describes him as “…a young turtle with a big imagination, Franklin tends to be very kind and lovable, and he has a large circle of friends”. No one could mistake Franklin for Pistol Pete. Whether Hegseth’s unauthorized use of Franklin is legal as a parody is unlikely but won’t be litigated. However, there is one parody that is all too evident. That is the spectacular unsuitability of Pete Hegseth, who reached the exalted rank of major in the National Guard and who clearly doesn’t know much about the role he is in, nor does he care, posing as the Secretary of Defense (or is it War?). That’s the true parody.

© Hugh Stephens, 2025. All Right Reserved

When the End Does Not Justify the Means, Anthropic’s $1.5 Billion Lesson

“Fair Use” Does Not Justify Piracy

A hand-written note on a white paper that reads 'END ≠ MEANS'.

Image: Author

The stunning announcement on September 5 that AI company Anthropic had agreed to a USD$1.5 billion out-of-court settlement to settle a class-action lawsuit brought by a group of authors was ground breaking in terms of its size, and goes to disprove the old adage that “the end justifies the means”. It is still not clear if the “end” (i.e. using copyrighted content without authorization to train AI algorithms) is legal, although preliminary indications are that at least in the US this may be the case. However, even if what Anthropic and other AI companies have been doing is ultimately determined to be fair use under US law—which is by no means certain—downloading and storing pirated content is clearly not legal, even if it is to be used for a fair use purpose. In other words, the piracy stands alone and must be judged as such, separate from whatever ultimate use to which the pirated content may be put.

Ironically, in the end, Anthropic did not even use much of the pirated content it had collected for training its platform, Claude. It seems to have had second thoughts about using content from online pirate libraries such as LibGen (Library Genesis) and PiLiMi (Pirate Library Mirror) and instead went out and purchased single physical copies of many works, disassembling and then digitizing them page by page for its Central Library, after which it destroyed the hard copies. Why go to all this trouble? Why not just access a legal online library? That’s because when you access a digital work, you don’t actually purchase it. You purchase a licence to use it, and that licence comes with conditions, such as likely prohibiting use for AI training. Anthropic would have been exposing itself to additional legal risk by violating the terms of the licence, so instead of negotiating a training licence, they took the easy way out by downloading content from pirate sites LibGen and PiLiMi. Later, having second thoughts, they purchased physical copies of the works they wanted to ingest and then scanned them. But it was too late. The piracy had already occurred.

When the decision in the Bartz v Anthropic case was released this summer, I commented that the findings were a mixed bag for AI developers. A very expensive mixed bag, it turns out. In the Anthropic case, there were clearly some interim “wins” for the AI industry. Anthropic’s unauthorized use of the works of the plaintiffs (authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson, who filed a class action suit) was ruled by the judge (William Alsup) to be “exceedingly tranformative” thus tipping the scales to qualify as a fair use. In addition, he ruled that Anthropic’s unauthorized digitization of the purchased books to also be fair and not infringing. However, it was the downloading and storing of the pirated works that got Anthropic into hot water. Even though the intended use of the pirated works was to train Claude, a so-called transformative fair use, this did not excuse the piracy. While Alsup did not specifically rule that use of pirated materials invalidates a fair use determination (i.e. he ruled that the piracy and the AI training were separate acts), his ruling exposes a weak flank for the AI companies. For example, the US Copyright Office has stated that the knowing use of pirated or illegally accessed works as training data weighs against a fair-use defence. In short, the end does not justify the means.

The piracy finding was significant because Judge Alsup decreed that this element of the case would be sent to a jury to determine the extent of damages. (In Canada and the UK, judges rather than juries normally play this role). Given that under US law statutory damages start at $750 for each work infringed but can go up to $150,000 per work for willful infringement, Anthropic could have been on the hook for tens of billions of dollars in damages for the almost 500,000 works at issue. (Over 7 million works were inventoried by the pirate websites and downloaded by Anthropic but the limitations on who qualifies for the class action reduced the number of actionable works to just 7 percent of the total). As deep as its pockets are (Anthropic is backed by Amazon), if a jury awarded damages toward the higher end of the scale, the company could have been bankrupted.

Thus, Anthropic had lots of incentive to settle (including keeping the fair use findings unchallenged). As it stands, the $1.5 billion payout, while large in total, amounts only to about $3000 per infringed work, not the minimum but not really financially significant for the plaintiffs. This amount will probably have to be split between authors and publishers, with some of the funds covering costs, so no authors are going to be buying a new house on the proceeds. The real beneficiaries will be the law firms that represented them. The messy process of deciding who gets what that has led Judge Alsup to suspend the proposed settlement in its current form and require greater clarity as to how the payouts will be managed. The number of works eligible for payment is limited by the fact that to qualify they have to meet three criteria;

1) they were downloaded by Anthropic from LibGen or PiLiMi in August 2022

2) they have an ISBN or ASIN (Amazon Standard Identification Number) and, importantly,

 3) they were registered with the US Copyright Office (USCO) within five years of publication, and prior to either June 2021 or July 2022, (depending on the library at issue).

Any other works do not qualify. Registration with the USCO is not a requirement for copyright protection but in a peculiarity of US law, without registration a copyright holder cannot bring legal action in the US.

While the settlement has been welcomed in copyright circles, and could set a standard for settlement in other pending cases where pirated material has been downloaded for AI training by companies such as META and OpenAI, it doesn’t settle the overriding question of whether the unauthorized use of non-pirated materials for AI training is legal. With the settlement, the Anthropic case is closed, including with respect to the fair use findings. There will be no appeal, another benefit for Anthropic. However, there are still a number of other cases working their way through the US courts, so the question of whether unauthorized use of copyrighted content for AI training constitutes fair use is far from settled.

The Anthropic settlement, especially its size, has caught people’s attention. It may result in AI developers deciding it is better to resort to licensing solutions to access content rather than risking the uncertain results of litigation. On the other hand, payments like this could be one-offs, a speed bump for deep pocketed AI companies who will continue to trample on the rights of creators if they can get away with it. In the Anthropic case, while the company must destroy its pirated database, it is not required to “unlearn” the pirated content that it ingested. Moreover, even if this case leads to more payments to authors, which would be welcome, there are still many copyright-related conundra to be resolved. It should not be necessary to have to constantly resort to litigation to assert creator’s rights given that, as the Anthropic case shows, only a very limited number of rightsholders benefit from specific cases. Broad licensing solutions are required. This would also help address the problem of AI platforms producing outputs that bear close resemblance to, or compete with, the content on which they have been trained.

While Bartz v Anthropic is a decision that applies only to the US, and only to this one very specific circumstance, it will be studied closely elsewhere in countries that do not follow the unpredictable US process of determining fair use, for example in fair dealing countries like the UK, Canada, Australia, New Zealand and elsewhere, and in EU countries. In Canada, the unauthorized use of copyrighted works for training commercial AI models is a live issue. With the possible exception of research, unauthorized use such as that undertaken by Anthropic is unlikely to fall into any of the fair dealing categories (in Canada, they are education, research, private study, criticism, review, news reporting, parody and satire) nor is there a Text and Data Mining (TDM) exception in Canadian law. As Canada and other countries come to grips with the copyright/AI training dilemma, the principle of how content is accessed will surely be an important principle. Just as fair use (if indeed AI training is determined to be fair use) does not justify piracy in the US, licit access is required in Canada to exercise fair dealing user rights, including where TPM’s (technological protection measures, aka digital locks) are in place to protect that content.

Judge Alsup’s decision upholds the important principle that the end (if legal) does not justify the means (if illegal). This is a key takeaway from the Anthropic case, imperfect as the outcomes of that case were. Meanwhile the legal process of determining how and on what terms AI developers should have access to copyrighted content to train their algorithms continues.

© Hugh Stephens, 2025. All Rights Reserved.

Hold the Champagne: The Two AI Training/Copyright Decisions Released in the US Last Week Were a Mixed Bag for AI Developers

Illustration of a champagne bottle being popped, enclosed in a red circle with a slash indicating 'no champagne'.

Image: Shutterstock.com

Last week I wrote about the questionable ethics of META’s use of pirated content to train its AI model, Llama, pointing out the ethical issues involved with META’s admitted use of pirated online libraries, such as LibGen (Library Genesis), to feed content to Llama for training purposes. This is quite apart from whatever legal issues that may arise from the widespread practice of ingesting copyrighted content for AI training by making an unauthorized copy from any source (such as a legitimate library, through purchase of a single copy of a work, or from publicly available internet sources, for example) not to mention the additional element of taking that content from pirate sources. The day after that blog was posted the first of what will be a series of legal decisions in the US regarding cases brought by authors and copyright holders against AI companies was issued, followed by another a day later. Both cases were heard in the Northern District of California, in the same San Franciso court house, but handled by different judges.

I updated last week’s blog to make reference to the Bartz v Anthropic case (hereafter “Anthropic”), but given the importance of that decision, combined with a decision released in another California court room a day later (Kadrey et al v META), these cases merit further exploration–especially since they were widely trumpeted by AI advocates as opening the door to unauthorized use of copyrighted content for AI training on the basis of “fair use”.

Fair use is the complex legal doctrine used in the US to determine exceptions to copyright protection. US readers are well aware of the intricacies and idiosyncrasies of fair use but for those not overly familiar with how it works, here is a short summation I drew from a blog post on fair use vs fair dealing that I wrote a few years ago.

In the US context, fair use is an affirmative defence against copyright infringement and is determined by the courts on a case by case basis, judged against several fairness factors (purpose and character of the use, the nature of the work copied, the amount and substantiality of the amount of the work used, and the effect of the use on the value of the original work)… Fair use is not defined by law. Some examples are given in US law of areas where the use is likely to be fair (criticism, comment, news reporting, teaching, scholarship, research) but these are illustrative and not exhaustive. In short, it is the courts that decide. This in turn can lead to extensive litigation as to what is and is not fair use, and it is worth noting that different judicial circuits in the US have at times come up with conflicting interpretations.

Or, for that matter, two different judges in the same circuit delivering decisions just days apart on similar issues but with some significantly different outcomes, as we saw last week (although in these cases both found fair use by AI developers with regard to the copyrighted works at issue).

On the Anthropic case, US District Judge William Alsup ruled, on summary judgement, that the use of copyrighted works for AI training, even though done without authorization, is highly transformative and does not substitute for the original work (“The technology at issue was among the most transformative many of us will see in our lifetimes”). It thus qualifies, according to Alsup, as fair use because the transformative nature of the use overrides or swallows the three other fair use factors, including the important fourth factor (effect of the use on the value of the work). He notes there was no allegation that the output of Anthropic’s model, known as “Claude”, produced content infringing the works of the plaintiffs. However, Judge Alsup then went on to consider the legality of Anthropic’s actions to download more than 7 million works from pirate libraries (such as Books3, Library Genesis and the Pirate Library Mirror) to constitute its reference library, which it initially planned to use for AI training. He concluded this was a prima facie case of copyright infringement, whether Anthropic intended to use some or all of the pirated works to train Claude or not. (“Anthropic seems to believe that because some of the works it copied were sometimes used in training LLMs (Large Language Models), Anthropic was entitled to take for free all the works in the world and keep them forever with no further accounting “.) Damages, to be decided at trial, could be substantial. Alsop did not, however, rule explicitly on whether or not the use of pirated works for AI training purposes could be a fair use.

Because of the controversial nature of Alsup’s findings on transformation and fair use, there is no question that this case will be appealed. While there have been many criticisms of the fair use elements of Alsup’s ruling, a particularly clear and trenchant analysis was put forth by Kevin Madigan of the Copyright Alliance (Fair Use Decision Fumbles Training Analysis but Sends Clear Piracy Message).

The second case last week to reach the decision stage was Kadrey et al v META. In this case District Judge Vince Chhabria found that META’s use of the works of the plaintiffs, thirteen noted fiction writers, to train its AI model (“Llama”) was also fair use. Chhabria, like Alsup, found that META’s use was transformative on the first fairness factor dealing with the purpose and character of the use (“There is no serious question that Meta’s use of the plaintiffs’ books had a “further purpose” and “different character” than the books—that it was highly transformative.”) but unlike Alsup, Chhabria put much greater emphasis on market harm, (the fourth fairness factor dealing with the effect of use on the value of the work) suggesting that it could be determinative. Unfortunately for the plaintiffs, however, Chhabria considered their arguments with respect to market harm to be unconvincing. There was no evidence that Llama’s output reproduced their works in any substantial way or substituted for the specific works at play nor was there evidence, according to the judge, that the unauthorized copying deprived the authors of licensing opportunities.

Chhabria suggested that a far more cogent argument would have been that use (unauthorized reproduction) of copyrighted books to train a Large Language Model might harm the market for those works by enabling the rapid generation of countless similar works that compete with the originals, even if the works themselves are not infringing. In other words, causing indirect substitution for the works rather than direct substitution. This is the theory of “market dilution”, which was also put forward speculatively by the US Copyright Office in its recent Pre-Publication Report on AI and copyright. Since this wasn’t presented as an argument, Chhabria could not rule on it but in effect he is inviting future litigants to pursue this line of argument, noting that his decision on fair use relates only to the works of the thirteen authors who brought the case.

The clearest way to illustrate his line of reasoning is to quote directly,

“In cases involving uses like Meta’s, it seems like the plaintiffs will often win, at least where those cases have better-developed records on the market effects of the defendant’s use. No matter how transformative LLM training may be, it’s hard to imagine that it can be fair use to use copyrighted books to develop a tool to make billions or trillions of dollars while enabling the creation of a potentially endless stream of competing works that could significantly harm the market for those books”.

This editorializing, known in legal circles as obiter dicta, is not binding nor precedential, yet will undoubtedly have some influence given Chhabria’s stature. It is likely that one of these days Judge Chhabria will have the opportunity to put these theories into practice when ruling on a similar case, but one where the plaintiffs have made a better case for market harm. He has provided them a roadmap.

While these two cases have fired the first shots in what is going to be a lengthy war, they do not seem to be dispositive. There are enough caveats and nuances to be able to conclude that the AI developers are far from being out of the woods. Both “victories” have a sting in their tail, especially Judge Alsup’s finding on piracy. Neither copyright advocates nor AI developers should be breaking out the champagne just yet. But whichever way it turns out, there will be some sure winners; the lawyers for each side.

© Hugh Stephens, 2025.

Is it Ethical to Use Pirated Content for Commercial Purposes? META Thinks So

Two signs hanging on a string, one labeled 'ETHICAL' in green and the other labeled 'LEGAL' in red, against a purple background.

Image: Shutterstock.com

There is the question of what is ethical, and then there is the question of what is legal. Sometimes they are the same, often not. The legality of using copyrighted content without authorization for commercial purposes, such as in training AI models—as META and a number of other companies have done—is being decided in court. In META’s case, however, there is the further complaint (not denied by META) that many of the unauthorized copies it made were taken from pirated content. While this revelation may not change the fundamentals of the copyright infringement case against it, there is still the ethical question for META to answer. On this, it comes up short. Very short.

META, the parent company of Facebook, Instagram and WhatsApp, used vast amounts of copyrighted content, without permission or licensing, to train its AI model. It is not alone in doing so. This practice may or may not be legal. A number of cases are working their way through the courts, most of them in the US, with copyright owners from Getty Images to Disney and Universal, from the New York Times to the Authors Guild and on to music labels, all claiming that their content was unfairly and illegally copied to provide training fodder for training AI models, such as META’s model, Llama. META and other AI developers claim that their use was a “fair use” under US law. We’ll see. However, as part of its giant vacuuming of publicly available (but in many cases protected) content, META also ingested content from various pirate sites and databases, notably the notorious “shadow library”, LibGen (Library Genesis). LibGen originated in Russia and contains up to 80 million scientific and academic articles, as well as millions of novels and nonfiction books, most unauthorized, unlicensed copies. It has a been sued by major academic and textbook publishers. In 2017 Elsevier won a $15 million judgement against LibGen, and another pirate website, SciHub. Last year Elsevier was awarded a $30 million default judgement. However, both LibGen and SciHub remain available online.

The extent of the copyrighted content held by LibGen was revealed in an investigative report published recently by The Atlantic. You can search through the LibGen database as published by The Atlantic to find out what works are included, and whether your work has been pirated. Authors from Newfoundland to New York and lots of places in-between and elsewhere found their works included in the database when they did the search. The Authors Guild advises writers to fight back by sending a formal notice to META and other AI companies asserting their rights, as well as adding a “No AI Training” notice on the copyright page of works. This is in addition, as would be expected, to joining the Authors Guild to help them fight what is happening.

Consuming pirated content can result in costly penalties, as some unfortunate downloaders have found out to their regret. Using it for commercial purposes is even more egregious. It’s like running a pirate streaming service based on stolen content. META didn’t use pirated content in this way, but they used it commercially just the same, in their case for AI training. Were they aware of what they were doing. You bet they were.

The discovery process in the US suit of Kadrey et. al. v META revealed a series of email exchanges in which some META employees expressed concerns over the ethics of using pirated content. The concerns went up the chain and back, with “MZ” (guess who? No, not Moses Znaimer) giving approval to proceed. Following on these revelations in Kadrey v META in the US, two class action lawsuits have been filed in Canada, one in Quebec on behalf of a number of French language authors and one in British Columbia. The Quebec suit specifically flags the piracy issue. Among the listed complaints is the following:

“Rather than acting within the law and respecting the rights of class members, it (META) deliberately chose to train its LLMs (Large Language Models) from datasets containing illicit copies of works from all over the world, including those of class members.”

Damages sought are $20,000 per work.

It is clear that META wilfully torrented content from LibGen, knowing that many or most of the works on LibGen were infringing, pirated copies. They just didn’t care.

If it turns out that somehow, inexplicably, META’s unauthorized use of copyrighted content for AI training is ruled by the US courts to be fair use, would the fact that the source of some of the content was from a pirate source be relevant? I am not sure, but a judgement that has just been delivered in California in the “Anthropic” case suggests that even if unauthorized copying can be justified as fair use because it is considered “transformative”, that does not excuse piracy–which is still an infringement. In this case, Anthropic copied both purchased and pirated works to train its AI model, and kept the copies in its central library. It was sued by some authors and journalists in a class action suit alleging copyright infringement. The judge, in one of the first such cases to reach a decision point, concluded on summary judgement that Anthropic’s unauthorized reproduction of copyrighted works for AI training was fair use under the transformation doctrine but added, with respect to those works drawn from pirate sources such as LibGen and others,

“piracy of otherwise available copies is inherently, irredeemably infringing even if the pirated copies are immediately used for the transformative use and
immediately discarded”.

A trial will be held to determine the damages from the piracy.

Being one of the first AI training cases out of the gate, Anthropic will certainly be appealed, so this is not the last word. However, this ruling when added to the US Copyright Office’s views expressed in its Pre-Publication Report on Generative AI issued last month on May 9, a day before Register Shira Perlmutter was dismissed, that “the copying of expressive works from pirate sources in order to generate unrestricted content that competes in the marketplace, when licensing is reasonably available, is unlikely to qualify as fair use”, suggests that META could be in both ethical and legal trouble.

In other jurisdictions, such as Singapore, content used in AI training under a Text and Data Mining exception has to be legally accessed, although this is very thin legal protection because technology companies can legally purchase just one copy of a work to comply. In Canada you cannot break the law (i.e. circumvent a technological protection measure) to exercise a fair dealing right. But whether or not using a pirated source puts META offside the law in the US with respect to fair use, (and the Anthropic case suggests that it could at least with respect to the pirated works), think of the ethics and the image this presents to the public.

A company like META, capitalized at something like $2 trillion, cannot be bothered to even access content legitimately, let alone use it legitimately. Why? Because MZ said it was ok to proceed. Sadly, even though they are not the only ones to use pirated content to train their AI models, that tells me all I need to know about the values and ethics of this particular company.

© Hugh Stephens, 2025. All Rights Reserved.

This post has been updated to include reference to the decision in the Anthropic case, released after the initial publication of this blog post.

More Balanced Interpretation of US “Transformation” Fair Use Test gives Rightsholders Better Leverage to Protect their Works

Photo: Author

A couple of recent cases in the US involving (mis)appropriation of copyrighted photographs indicate the tide may be changing with respect to the interpretation of what constitutes fair use, moving from a very liberal interpretation of “transformation” that has been used in recent years to justify unauthorized reproduction of copyrighted works by making them into “something (allegedly) new”, to a more balanced interpretation where the rights of the original creator/rightsholder are given greater weight. These cases are Andy Warhol Foundation v Goldsmith, decided by the US Supreme Court in May of 2023 and now, just last week, two cases (Graham v Prince and McNatt v Prince) in New York District Court. These involved photographers Donald Graham and Eric McNatt who had sued prominent appropriation artist Richard Prince and two well-known galleries, Gagosian and Blum & Poe, for copyright infringement. A final judgment was entered in these cases wherein Prince was found liable for willful infringement of the photographs, required to pay damages totalling five times the value of the infringing works he sold, plus paying the plaintiff’s legal costs, while also being restrained from reproducing, modifying, or distributing the contested artwork that was developed from the appropriated photographs. This is a big deal for photographers and other creators whose work is appropriated by others without authorization on the pretext of “transformation”. It could even have a ripple effect outside the US.

Last week my blog post was about photography, looked at through a historical lens from a copyright perspective. Photographers initially had a struggle to gain copyright protection for their works, at least in the US, but for many years now have enjoyed the benefits of such protection. However, just because photographs are recognized as copyright protected works doesn’t mean that infringement does not occur on a regular basis, or that photographers have not had their work appropriated without authorization as a result of court rulings interpreting what constitutes fair use (in the US).

For those readers not immersed in the minutiae of copyright, what follows is a brief explanation about fair use in the US and fair dealing in Canada. (Others may choose to skip to the next paragraph). Fair use is the provision in US law that allows users to reproduce parts of copyrighted works without permission or authorization from the rights holder, subject to various limitations and conditions. Unlike in Canada or a number of other countries where legal exceptions to copyright are based on a specified list of “fair dealing” purposes, the US fair use list is only illustrative. There is a four-part test in the US that courts use to judge whether an unauthorized use constitutes fair use. The four factors that are weighed are; (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. The “transformation test”, which has been an important factor in determining fair use in the US in recent years, comes in under the first category of the purpose and character of the use. Transformative use has been interpreted by US courts to mean the addition of new expression or meaning to an original work to create something new with a different purpose. Parody, which is both a fair dealing exception in Canada and one of the illustrative fair use categories in the US, is such an example where the original work is transformed to give it a new and different meaning. But transformation in the US can involve much more than just parody. The 2015 Google Books case (Authors Guild v Google Inc) where Google was sued for copyright infringement for unauthorized scanning and display of copyrighted works was decided, in Google’s favour, on the basis of “transformation”. The US Court of Appeals found that, “Google’s making of a digital copy to provide a search function is a transformative use, which augments public knowledge by making available information about Plaintiffs’ books without providing the public with a substantial substitute for matter protected by the Plaintiffs’ copyright interests in the original works…”. Notably Google, in settling an earlier case with US publishers, had agreed to limit the public display of a copied work to no more than 20 percent of its contents. The transformative use argument has been used as a fair use defence in a number of US copyright cases, including some involving photography. In these cases, the entire photograph was taken without authorization but then “transformed” by the addition of some new content.

One notable such case is Cariou v Prince, finally decided on appeal in 2013. In this case, Richard Prince successfully claimed fair use for his adaptation of photographs taken by French photographer Patrick Cariou. Prince did a series of paintings that incorporated photos of Rastafarians taken from Cariou’s book “Yes, Rasta”. The Appeal Court found that Prince’s paintings were not only sufficiently transformative to be different works, but they did not compete with the original in the marketplace. This case was part of a trend where US courts successively widened the interpretation of fair use based on a liberal application of the transformation principle.

The antidote to Cariou is the more recent Warhol v Goldsmith case, mentioned above. In this case, celebrity photographer Lynn Goldsmith sued the Warhol Foundation for copyright infringement in 2017 after she learned that her 1981 photograph of the noted singer-songwriter-musician Prince, who died in 2016, (not Richard Prince the artist, just to confuse things), had been used by Warhol (who himself died in 1987) to create a series of 14 silk screen images back in 1984. At the time, Goldsmith had been paid $400 by the publisher Conde Nast for a one time use of her photograph as an artist’s reference for Warhol to create an image for a Vanity Fair article about Prince, Purple Fame, published in November 1984. However, she was unaware that Warhol had created a whole series of prints from her photo. Years later, Vanity Fair paid the Warhol Foundation $10,000 for use of another of the prints in the series, Orange Prince, in an article to mark Prince’s death. Goldsmith was neither paid nor credited. She sued (actually countersued) for infringement after the Foundation sought a declaratory judgment that the creation of the series was fair use. It was an uphill battle for Goldsmith against a very wealthy and well-known Foundation, but in the end she won. The US Supreme Court decided in May of 2023 that Goldsmith’s copyright had been infringed—but only in the one instance regarding the use of Orange Prince, which was seen as a market substitute for Goldsmith’s original photograph. The ruling was 7-2 with Justice Sotomayer writing “Goldsmith’s original works, like those of other photographers, are entitled to copyright protection, even against famous artists”. 

The doors of transformative use, swinging wide open for the past couple of decades, had just closed a bit. It is not irrelevant to note that the transformative use argument is the main fair use defence put forward by AI developers to justify their unauthorized scraping and ingestion of copyrighted content to train generative AI algorithms. Now, on the heels of the Warhol v Goldsmith decision, we have the outcome of the two photographer’s cases against Richard Prince and the galleries. This suggests that the seemingly open-ended “transformation” standard in the US is changing, or at least that the Supreme Court’s Warhol decision is sending a message that open season on photographers (and other creative works) by appropriation artists may be coming to an end.

One of the two photography cases against Richard Prince goes back almost eight years, to about the time that Prince prevailed in the Cariou case. Graham, whose 1998 photograph “Rastafarian Smoking a Joint” was appropriated by Prince, launched his suit in 2016. McNatt’s suit, for infringement of his photograph of musician Kim Gordon, came four years later. Last year Richard Prince tried to get the cases dismissed, but the judge refused. Prince had added very little to the photographs, mostly some gibberish text, yet the “artworks” were sold for substantial sums and exhibited in the prestigious Gagosian Gallery, a co-defendant.

The cases did not go to trial by jury. On January 25, a final judgement was entered by the parties and accepted by the court in which Prince agreed to pay the damages, which were far in excess of what statutory damages would have been. While Richard Prince’s attorneys are claiming that the artist did not admit infringement, the fact that he decided to avoid going to trial and agreed to the damages is a pretty realistic appraisal of what the likely outcome of a trial would have been. Unlike a “settlement”, which the two photographers refused, a “final judgement” is not subject to a confidentiality agreement as to the terms. According to legal counsel representing the photographers, they rejected a confidential settlement (where damages could have been higher) because they wanted the appropriation issue exposed. They succeeded.

It is impossible to know precisely what impact the Warhol/Goldsmith case had on this outcome, but it is surely not coincidental that after almost eight years of litigation (in the case of plaintiff Graham) that Richard Prince decided to put this behind him and cut his losses just months after the Supreme Court’s decision. After Warhol, the Prince case marks another example of some restoration of balance to the interpretation of transformative use. It is not enough to simply add a few features to an original work, and claim it is somehow “transformed” into something entirely different. The lawyer representing the photographers described this as a David versus Goliath case, where a couple of individual photographers took on a major artist and gallery–and won. This demonstrates that fair use as applied in the US clearly has limits and that even the rich and famous can be called to account.

It remains to be seen whether this trend, which started with Warhol, will continue but it is a positive sign for artists wanting to protect their work from unauthorized reproduction and appropriation. The Warhol case has no direct legal impact outside the United States, of course, but it has been taken note of elsewhere. It could mark a turning of the tide. Proponents of further widening copyright exceptions in Canada by modifying fair dealing criteria to be illustrative, as in the US, rather than prescriptive as is currently the case, (by inserting “such as” when describing fair dealing categories) will take little comfort from this. The false allure of more permissive unauthorized copying propped up by the US transformative use doctrine has already seeped across many borders. Now perhaps the counter trend will also spread.

With respect to the Warhol case, one Canadian IP lawyer has stated “I think SCOTUS’s application of fair use to the facts, in this case, will be persuasive to Canadian courts when they have to apply fair dealing to similar facts.” While the transformation doctrine does not directly apply in Canada, it has some applicability with respect to parody. We shall see in time what the long-term impact may be. But from the perspective of creators everywhere, reinserting some much-needed balance into the US fair use transformation standard is a welcome and important development as copyright law continues to evolve.

© Hugh Stephens, 2024.