CAI Policy Insights: GenAI Legal Update Edition
Welcome to CAI Policy Insights, a monthly policy digest covering the latest legal and policy updates impacting online and hybrid learning and the use of educational technologies. By staying up to date on news and emerging controversies in these areas, we believe faculty, administrators, learning experience designers, and academic leaders can all make more informed decisions regarding program development, technology integrations, student engagement and assessment strategies, and more. Topic-by-topic breakdowns of key regulatory issues can also be found on the Online Teaching Compliance Page.
Summary and Insights
Recent court cases over artificial intelligence training data have significant implications for online education. If the law ultimately requires licensing for training content, AI systems being used by institutions may become expensive and, in some instances, inaccessible, especially for smaller or less resourced institutions. Courts are split on whether using copyrighted materials to train generative AI is “fair use.” In Thomson Reuters v. Ross, using copyrighted materials for AI training was not considered fair use, while other decisions (Bartz v. Anthropic and Kadrey v. Meta) did find using the material for training to be a fair use.
In addition, the U.S. Department of Education has started the process of promulgating new implementing regulations for the One Big Beautiful Bill Act (OBBBA), moving forward with reduced resources during the current government shutdown. Meanwhile, the department has successfully defended Biden-era accountability frameworks (i.e., the new Financial Value Transparency framework and updates to Gainful Employment rules) in court, which will now need to be harmonized with OBBBA’s new accountability framework. Each accountability framework may impact online programs in unique ways based on how students located outside of an institution’s home will be represented in the various return-on-investment calculations associated with each framework.
Feature Policy Update: Generative AI Case Updates
There are currently over 50 pending copyright lawsuits against AI companies that focus on whether training AI systems on copyrighted content is a “fair use.” Most of these cases involve GenAI systems that use training data and inputs to create new works based on user prompts. Some of the first cases have had major movement this year, giving us a peek into what the future holds for these kinds of tools.
Thomson Reuters Enterprise v. Ross Intelligence
Thomson Reuters owns the legal research platform Westlaw. Ross took the headnotes—short summaries of key legal findings in cases offered by Westlaw—and used them to train their AI system. The court found that the headnotes were copyrightable even though they were summaries of uncopyrightable judicial opinions, and that Ross’s use of them in training was not a fair use.
Ross also argued that if their use wasn’t a fair use, their use of the headnotes was covered by the merger doctrine. The merger doctrine comes into play when factual or non-copyrightable information can only be conveyed in one way. In those instances, that one way of expressing the information is also uncopyrightable.
The merger doctrine applies because Ross didn’t directly take the headnotes from Reuters. They hired a third party to write answers to legal questions they thought their AI system would be asked. The third party then wrote some of the summaries themselves, sometimes using the Westlaw headnotes for reference when they contained quotes from the judicial opinions that were necessary for the summary. Ross said that the portions of the headnotes used in the training were uncopyrightable because they were the only way to convey the uncopyrightable legal findings. The court was not convinced.
Bartz v. Anthropic
Anthropic used millions of in copyright books to train their GenAI system, Claude. Bartz, who represents a class of authors whose works were part of the training data, sued Anthropic for two things: (i) using their books to train Claude and (ii) using their books to create an internal-facing library of all the works in existence.
The court found that the use of the books to train Claude was a transformative fair use because the system only used the works to learn; it did not generate the text of the authors’ works for end users. It also found that the books Anthropic legally acquired (through purchasing used physical copies and digitizing them) could be included in the library.
But the works that were acquired illegally through shadow libraries (i.e., sites where books, journal articles, and other content can be downloaded for free) were not allowed to be stored in Anthropic’s library. Since the pirated works were not used to train Claude or any other LLM, it was not decided whether or not the pirated status of the work would affect the fair use analysis for training. Instead of going to trial over the pirated books, Anthropic recently settled the case for $1.5 billion; so this issue remains unresolved until or unless another court addresses it.
Kadrey v. Meta
Similar to Anthropic, Meta was sued by book copyright holders for using their books to train LLAMA, Meta’s GenAI system. As in Bartz, the court decided on summary judgment that Meta’s use of the books was a fair use. The fact that Meta torrented the books from shadow libraries did not impact the court’s finding that the use was fair.
In this case, the court introduced a new concept to the fair use analysis: market dilution. Under the market dilution theory, if a use would oversaturate the market and reduce profits for the original work, the use will not be fair, even if it’s transformative. Because the authors didn’t present any evidence addressing market dilution, this new theory had no impact on the fair use analysis. However, the court did outline how future plaintiffs could successfully argue this theory in other cases.
Key Takeaways
The outcomes of these cases may have massive implications on the future of GenAI, including how and when these tools can be used in online learning. And with two cases decided for fair use but one against, it’s still unclear how the rest of the AI companies will do in the remaining lawsuits (especially since Ross is being appealed and Kadry likely will too).
In a potential future where AI software companies need to license the millions of pieces of content needed to train their systems, the cost of accessing these systems could go up dramatically. Innovation may slow, with smaller companies, in particular, potentially finding it too resource-intensive to design new tools given the potential for added licensing costs. If high licensing costs limit companies to a smaller or less inclusive set of training data, it will result in lower-quality GenAI outputs that are more biased.
Because fair use is so fact-specific, it’s unlikely that a verdict will come out that will easily solve all the cases, unless, of course, one of the cases makes it to the Supreme Court. Until then, we’ll continue to watch pending cases and keep you updated when there are important developments, so continue to watch this space!
Other Developments Worth Monitoring
Other major policy developments that could potentially impact digital learning are outlined below.
- Proposed Changes to SARA Policies. The 2025 Policy Modification Process (PMP) for the State Authorization Reciprocity Agreements has reached its final stage, with nine proposals receiving unanimous approval by regional steering committees. The final step in the PMP is for the board to offer a final vote, which will be available to the public via a livestream broadcast on October 23, 2025, at 2 p.m. Eastern Time. Coverage and registration link from NC-SARA.
- Proposed Changes to IPEDS Reporting. The Education Department is proposing revisions to the Integrated Postsecondary Education Data System (IPEDS) that would add required admissions reporting and transparency requirements to more directly tie applicant race and sex data to admission test performance, GPAs, and family income as part of implementation efforts consistent with recent executive orders and memorandums from the Trump administration regarding the elimination of diversity, equity, and inclusion practices in higher education. The comment period recently concluded on October 14, 2025. Link to proposed changes and comment portal.
- More Education Department Staffing Updates. After a federal district court had again compelled the department to bring back Office of Civil Rights employees following a massive reduction in force in the spring that had been temporarily approved to go forward by the U.S. Supreme Court, an appellate court has now overturned the lower court’s order. This will halt plans from the department to bring back employees throughout the fall and allow for the initial terminations to remain in place. With the civil rights office operating at about half capacity compared to recent years, critics of these terminations have warned about major challenges that the office will face in effectively enforcing critical civil rights protections in higher education (including for students participating in online education programs). Coverage from Higher Ed Dive. Relatedly, with the current government shutdown, the department has also released its contingency plan, which would involve a significant percentage of furloughed employees—an approximately 95% staff reduction—across ED after the first week.
- New H-1B Fee and Immediate Legal Challenges. After the Trump administration announced a new $100,000 fee for H-1B visas, a number of higher education groups are now suing the Trump administration to stop enforcement. Coverage from Higher Ed Dive.
- Gainful Employment Case Dismissed. Following the department’s unexpected defense of the Biden-era Financial Value Transparency and Gainful Employment rules in court over the summer, a federal court has now dismissed consolidated cases where plaintiffs argued the new and expanded accountability measures and formulas used to determine whether programs provide satisfactory economic returns on investments for students violated the Administrative Procedure Act. Upcoming negotiated rulemaking from the department, scheduled to begin in December, may provide clarity on how these frameworks will now work with new, similar accountability requirements found in OBBBA. Coverage from Higher Ed Dive.
Past Editions
Looking for news that first broke in a prior month or perhaps for historical context of a story featured in this article? Links to past editions of CAI Policy Insights are provided below.
Accessibility Coordinator Caroline Damren and Associate Director, Compliance and Policy Ricky LaFosse contributed to this report.
