 |
| Daniel J. Escott |
While the ruling is grounded in American copyright law, its implications extend far beyond American borders, particularly for Canada, where copyright protections and AI governance are at a pivotal moment.
The decision: Fair use, copyright, and AI training
At the core of the U.S. court’s decision was the question of whether ROSS’s use of Westlaw
nicoletaionescu: ISTOCKPHOTO.COM
1. Thomson Reuters’ headnotes were copyrightable: The court found that Westlaw’s headnotes, which summarize case law, contain sufficient originality to be protected under copyright law.
2. ROSS copied a substantial portion of the work: The evidence showed that ROSS used thousands of Westlaw headnotes to develop its AI’s capabilities, leading the court to determine that substantial copying occurred.
3. The use was not transformative: ROSS argued that its AI transformed the content into a new product, but the court found that the tool functioned similarly to Westlaw’s search and categorization system, making it a competing substitute rather than a transformative innovation.
2. ROSS copied a substantial portion of the work: The evidence showed that ROSS used thousands of Westlaw headnotes to develop its AI’s capabilities, leading the court to determine that substantial copying occurred.
3. The use was not transformative: ROSS argued that its AI transformed the content into a new product, but the court found that the tool functioned similarly to Westlaw’s search and categorization system, making it a competing substitute rather than a transformative innovation.
4. Market Harm to Thomson Reuters: The court emphasized that ROSS’s AI tool could compete with Westlaw, diminishing the value of Thomson Reuters’ copyrighted materials and negatively impacting its market.
This decision underscores the increasing judicial scrutiny AI developers face when using proprietary datasets for training models. The implications of the ruling extend beyond legal research tools, touching on AI development in various industries reliant on large datasets.
Implications for Canadian copyright law
Canadian copyright law, while distinct from U.S. law, shares similarities in its approach to originality, fair dealing (akin to fair use), and database protection. The Thomson Reuters ruling raises important questions for Canadian copyright jurisprudence, particularly in how Canadian courts might address AI training data and fair dealing exceptions.
1. Fair dealing and AI training in Canada
Unlike the U.S., Canada does not have an explicit “fair use” doctrine but instead applies fair dealing, which has narrower statutory exceptions. Under the Copyright Act, RSC 1985, c C-42, fair dealing is permitted for purposes such as research, private study, criticism, and education. If a Canadian AI company were to train its model using copyrighted materials, it would need to argue that its use falls within these exceptions.
However, the recent U.S. ruling suggests that courts may view AI training as a commercial and competitive use, rather than research or private study. If Canadian courts adopt a similar reasoning, AI companies may find it difficult to justify the unlicensed use of proprietary materials for training datasets.
2. Database and compilation protection
The U.S. court’s recognition of Westlaw’s headnotes as copyrightable compilations aligns with Canada’s protection of databases. Canadian courts have previously upheld copyright in compilations, particularly where editorial discretion or selection demonstrates originality. The Thomson Reuters ruling reinforces the notion that AI developers cannot freely scrape or extract structured data from proprietary sources without permission.
This raises significant concerns for AI companies that rely on data aggregation. In Canada, industries that depend on legal, financial, or medical databases may see tighter restrictions on how they access and use proprietary datasets for AI training.
3. The Potential for AI-Specific copyright exemptions
Given the growing reliance on AI across industries, policymakers may need to reconsider whether copyright law should evolve to include AI-specific exemptions for data use. Some jurisdictions, including the EU, have introduced text and data mining exceptions that allow AI developers to use copyrighted materials under certain conditions.
In Canada, the discussion around AI and copyright law is still evolving. The federal government has launched consultations on AI governance, including copyright and intellectual property considerations. The Thomson Reuters ruling could serve as a catalyst for Canadian policymakers to clarify whether AI training should be accommodated under fair dealing or require explicit licensing agreements.
Broader implications for AI development
Beyond legal research, this case signals a significant shift in how courts may regulate AI training data. Companies developing AI tools for natural language processing, medical research, financial analytics, and even creative industries (such as generative AI models for art and music) will need to assess the risks associated with using copyrighted materials.
Some key takeaways include:
- Increased Scrutiny of AI Training Data: AI developers must carefully consider whether their training datasets contain copyrighted materials and whether they have obtained proper licensing agreements.
- Rising Costs for AI Development: As companies face legal risks for unauthorized data usage, they may be forced to rely on licensed datasets, potentially increasing the cost of AI research and development.
- Shifts Toward Open Data Models: The legal uncertainty surrounding proprietary data may encourage the use of open-source legal datasets, government repositories, and publicly available court decisions to train AI systems.
A defining moment for AI and copyright law
This decision is one of the most consequential rulings on AI and copyright law to date. It establishes that the unlicensed use of structured, copyright-protected content to train AI tools can constitute infringement. While this ruling was issued in the United States, its impact will undoubtedly shape global AI governance, including in Canada.
As AI continues to disrupt traditional industries, the balance between innovation and intellectual property protection will become a defining legal issue. Canadian courts and policymakers must now grapple with the complex reality of AI-driven content creation and the ethical use of data. Whether through stricter enforcement of existing copyright laws or the introduction of new AI-specific exemptions, the conversation surrounding copyright, fair dealing, and AI training is only just beginning.
For AI developers, legal professionals, and policymakers alike, this case serves as a wake-up call: the era of unchecked AI data scraping is over, and the future of AI innovation must align with evolving copyright standards.
Daniel J. Escott is a research fellow at the Artificial Intelligence Risk and Regulation Lab and the Access to Justice Centre for Excellence. He is currently pursuing an LL.M. at Osgoode Hall Law School, and holds a J.D. from the University of New Brunswick and a BBA from the Memorial University of Newfoundland.
The opinions expressed are those of the author(s) and do not necessarily reflect the views of the author’s firm, its clients, Law360 Canada, LexisNexis Canada or any of its or their respective affiliates. This article is for general information purposes and is not intended to be and should not be taken as legal advice.
Interested in writing for us? To learn more about how you can add your voice to Law360 Canada, contact Analysis Editor Peter Carter at peter.carter@lexisnexis.ca or call 647-776-6740.