TLDR: Recent summary judgments in the Anthropic and Meta cases indicate a growing recognition of generative AI’s transformative nature, generally favoring fair use for copyrighted material used in training large language models (LLMs). While these rulings offer some clarity, concerns regarding unauthorized data harvesting and potential market disruption for original works remain key areas of contention in ongoing litigation.
The legal landscape surrounding generative artificial intelligence (GenAI) and copyright fair use is rapidly evolving, with recent summary judgments in cases involving Anthropic and Meta Platforms Inc. providing significant insights. These rulings, particularly from the Northern District of California courts, highlight the ‘highly transformative’ character of GenAI, which is a crucial factor in determining fair use for copyrighted materials used in training large language models (LLMs).
In the case of Bartz et al. v. Anthropic PBC, Judge William Alsup issued a summary judgment on June 23, 2025, concluding that Anthropic’s LLM was ‘quintessentially transformative’ in its use of training material. This decision centered on the principle that the act of training an AI model, even with copyrighted works, can be considered transformative if the AI’s output does not directly reproduce the original work but rather generates new content based on the learned patterns. However, Judge Alsup’s ruling also distinguished between legitimately purchased copyrighted material and content obtained from ‘pirate sites on the internet,’ allowing claims related to the latter to proceed to trial to ascertain damages.
Similarly, in Kadrey et al. v. Meta Platforms Inc. et al., Judge Vince Chhabria, in a decision filed on June 25, 2025, also found the output of Meta’s GenAI platforms to be transformative. Despite this, Judge Chhabria expressed concerns about the potential for GenAI to create works that could directly compete with or substitute for authors’ original creations, thereby impacting their market value. While the fair use defense applied in this specific case due to the hypothetical nature of the harm presented, the ruling underscored that future cases with better-developed records on market effects might yield different outcomes.
These decisions underscore the four factors typically considered by U.S. courts when assessing fair use: (i) the purpose and character of the use, particularly its transformative nature; (ii) the nature of the copyrighted materials (factual works are more likely to qualify for fair use than fictional ones); (iii) the amount and substantiality of the portion used; and (iv) the effect of the use on the potential market for or value of the copyrighted work. The ‘transformativeness’ of GenAI has been a dominant theme in these recent judgments, often outweighing other factors in initial rulings.
However, the legal battles are far from over. The United States Copyright Office (USCO), in a pre-publication report from June 9, 2025, indicated that the application of fair use to generative AI is nuanced, stating, ‘It depends.’ The USCO emphasized the importance of the first and fourth factors, noting that training an AI for research purposes is likely transformative, while training it to generate content directly imitating copyrighted works is not. The USCO also highlighted concerns about market dilution, lost sales, and lost licensing opportunities for human artists and writers due to AI’s ability to create competitive content or imitate styles.
Also Read:
- Federal Judge Rules in Favor of Meta in AI Copyright Lawsuit, Citing Flawed Arguments by Authors
- Anthropic Faces Multi-Billion Dollar Copyright Lawsuit Over Alleged Use of Pirated Books in AI Training
Previous cases, such as Thomson Reuters versus Ross Intelligence earlier this year, have shown that fair use exceptions are not universally applicable, especially when the AI’s use is not considered transformative and directly competes with the plaintiff’s works. The ongoing litigation will continue to shape how courts interpret and apply the fair use doctrine in the context of rapidly advancing AI technologies, with particular attention to how AI companies acquire their training data and the demonstrable impact of AI-generated content on existing markets.
