In yet another battle with the use of copyrighted materials to train AI models, I came across this interesting news article[1], which caught my attention about copyright law professors in the US submitting an amicus brief[2] in support of authors suing Meta in the case of Richard Kadrey, et al. v. Meta Platforms, Inc. for allegedly training its Llama AI models on e-books without permission and intentionally removing the copyright information from those e-books to hide the alleged infringement. I will be discussing the summary of the brief and fair dealing provisions in Canada’s Copyright Act, as opposed to the similar yet distinct fair use provisions found in United States copyright law.
Summary of the brief
The copyright law professors take the position that the Meta’s fair use claim is a “breathtaking request for greater legal privileges than courts have ever granted human authors.” According to them, the use of copyrighted works by Meta to train generative AI models is not “transformative” because it’s not different from using them to educate human authors, which is an original purpose of author’s works. Also, such training of Llama AI models is creating the works that would compete with the copied works in the same market place. The use of works is also not a “non-expressive use,” because it incorporates the expressive choices of the authors of those works into the models. Lastly, they argue that the measure of actual damages for copyright infringement does not include the value of skills learned from infringing copies, thus, limiting damages is crucially different from granting immunity from all liability. For these reasons, the case for fair use is weak.
Is Copying by Meta, Fair Dealing under the Canadian Law
The fair dealing provision in the Copyright Act (Section 29) permits use of a copyright-protected work without permission from the copyright owner or the payment of copyright royalties. To qualify for fair dealing, two tests must be passed. First, the “dealing” (or copying) must be for a purpose stated in the Copyright Act i.e. research, education, private study, criticism, review, news reporting, satire, and parody. The second test is that the dealing must be “fair.”[3] The Copyright Act does not define what will be “fair”, whether something is fair is a question of fact and remains a case-by-case analysis.[4] Because fair dealing is a fairly broad, open and flexible device, significant court decisions have provided some useful considerations and parameters.
The most important is the Six Factor Analysis articulated in the landmark copyright case CCH Canadian Ltd. v. Law Society of Upper Canada[5]. To determine whether or not copying is Fair Dealing, the court suggested six non-exhaustive factors to be considered: purpose, character, amount, alternatives, nature, and effect. The onus is on the user invoking “fair dealing” to meet the requirements of the test[6]. Based on these guidelines, sectors such as education, journalism and film production have developed best practices for applying fair dealing.
Expanding on this discourse, the Court in SOCAN v Bell[7] stated that fair dealing is an essential part of furthering the public interest objectives of Canada’s copyright legislation, and in helping to achieve the balance between copyright protections, and public access to works, the fair dealing provision “must not be interpreted restrictively”. The purpose of “Fair dealing” exception is to strike a balance between creators having exclusive rights over their original creations and allowing the general public to benefit from those creations without incurring undue financial hardship. Any amount of excessive control by the authors may result in unduly limiting the ability of the public domain to embrace creative innovation.
Applying the defence of fair dealing if Meta would have argued in Canada, as established in CCH Canadian Ltd v Law Society of Upper Canada, –
Under the first step of the analysis, Meta could argue that its use of copyrighted materials for training its Llama AI models fall under the purpose of education and research. The second step of the analysis must look at the following factors:
- Purpose of the Dealing: While Meta could argue that the purpose was to create e-books for educational and research purposes, but for the tech-giant like Meta, also makes the use undeniably commercial.
- Character of the Dealing: This factor looks at how the works are dealt with, for example are multiple copies widely distributed, is a single copy made. In this matter, Meta has copied author’s works in their entireties and is creating competing product, thus might be seen as less fair.
- Amount of the Dealing: This factor looks at the quantity and quality of the infringing use. Since the work copied by Meta is extensive, it would likely be seen as excessive, and would weigh against fairness.
- Alternatives to the Dealing: It will be hard to imagine non-infringing alternatives for the author’s original creations in this context.
- Nature of the Work: Meta had used copyrighted books to train Llama AI models, which courts are inclined to protect.
- Effect of the Dealing on the Work: The Meta’s product will directly compete with author’s original works, thus would weigh against fairness.
In view of the above, it is unlikely that fair dealing exemption would apply under the Canadian law.
Moving Forward
Thomson Reuters Enterprise Centre GmbH et al v. ROSS Intelligence Inc.[8] has been the first major U.S. AI-copyright decision that answered the question of whether the fair use defence protects an AI model provider from a copyright infringement claim. The outcome of this case may differ from other outstanding AI-copyright cases such as the present case of KADREY, et al. v. META. Although Thomson Reuters v. ROSS Intelligence did not involve generative AI, but it is still an important decision because it laid down the foundational bricks in the copyright jurisprudence considering AI and fair use.
Having said that, it is important to note that fair use or fair dealing analysis are fact specific and in the AI context, it will depend on the differences between purposes, type of content used, training process, and outputs. Before businesses like Meta can embrace the benefits of generative AI, they need to understand the risks and how to protect themselves.
Conclusion
Overall, I would assume that battles against companies creating AI tools will happen quite frequently raising serious copyright infringement consequences, especially when legal scholars await an important precedent from Canadian courts regarding how to handle copyright issues in the context of AI and further providing valuable guidance for the technology sector.
[1] https://techcrunch.com/2025/04/11/law-professors-side-with-authors-battling-meta-in-ai-copyright-case/
[2] https://storage.courtlistener.com/recap/gov.uscourts.cand.415175/gov.uscourts.cand.415175.525.0.pdf
[3] CCH Canadian Ltd. v. Law Society of Upper Canada 2004 SCC 13 at para 50
[4] Ibid at para 52
[5] CCH Canadian Ltd. v. Law Society of Upper Canada 2004 SCC 13
[6] Alberta (Education) v. Canadian Copyright Licensing Agency (Access Copyright) 2012 SCC 37at para 12
[7] Society of Composers, Authors and Music Publishers of Canada v. Bell Canada 2012 SCC 36 at para 11
[8] https://www.ded.uscourts.gov/sites/ded/files/opinions/20-613_5.pdf
Copyright & Social Media
Communications Law