The Implications of US Court Rulings on Fair Use of AI Training Models for Companies in Israel

At the end of June 2025, the Northern California District Court issued two precedent-setting rulings addressing the legal question: Is it lawful to use copyrighted works for machine training of AI-based large language models (LLMs)?

The court rulings relate to two significant players in the market, Anthropic and Meta. In both lawsuits, the court allowed, in principle, the “fair use” defense under US law (which also exists in Israeli law). These rulings join a series of similar high-profile lawsuits adjudicating similar legal questions we’ve reviewed in the past, including the New York Times lawsuit against OpenAI and Microsoft, the lawsuits filed against Midjourney and Stability AI, and the Thomson Reuters lawsuit against Ross Intelligence.

Bartz v. Anthropic: Claude – an AI Model That Mimics Human Learning

Three authors alleged that Anthropic (which develops generative AI models) copied over seven million copies of their works from pirated and purchased sources to train its LLM, Claude. Anthropic only admitted to using purchased copies. Anthropic also admitted that it holds pirated copies in a separate collection, but claimed it does not use the pirated copies for training purposes.

The court examined the four “fair use” tests under the US Copyright Act, which Israel has also adopted for its own fair use doctrine: the purpose and character of the use, the nature of the copyrighted work, the amount and substantiality of the portion used in relation to the copyrighted work as a whole, and its impact on the potential market for or value of the copyrighted work.

The court found that most of the considerations tilt in favor of Anthropic, mainly because its training process is transformative – similarly to how people learn from books – and as a source of inspiration, rather than copying or distributing the copyrighted work itself. Therefore, the court ruled that the use of the purchased copies constitutes fair use.

However, the court ruled that possession or use of pirated copies does not excuse Anthropic’s piracy and that Anthropic is liable for theft. The scope of Anthropic’s liability for theft will be adjudicated in a separate proceeding.

This class action could expose Anthropic to damages at the volume of up to USD 150,000 per copyrighted work.

Kadrey v. Meta: No Harm, No Foul

Two days after the Anthropic case ruling, the same court granted Meta’s motion to dismiss a similar lawsuit filed against it. In this case, the court ruled that the use of copies originating from pirated libraries may be considered fair use, because the authors failed to present evidence of dilution of the market for their copyrighted works. As stated, the fourth fair-use test – harm to the market for a copyrighted work – is probably the most important, since the purpose of copyrighting is to protect authors’ economic rights to revenues and prevent copyright infringement. In this case, since the plaintiffs failed to present material evidence of dilution of the market for their copyrighted words, the lawsuit was dismissed.

Nevertheless, the judge clarified that his ruling was based on the specific facts of the case, without ruling whether Meta’s actions were lawful per se. Therefore, future adjudications of similar cases could result in a different outcome if plaintiffs submit solid evidence of dilution of the market for their copyrighted works.

Not the Final Word on Fair Use of AI-Based Models

These two court rulings constitute the first milestone in the ongoing efforts to strike a balance between copyright protection and the training needs of AI systems at the vanguard of technological progress in making content accessible to the general public.

On the one hand, the ruling in the Anthropic case opens the door to transformative uses of AI being protected under the fair use doctrine, i.e., uses of copyrighted works for AI training purposes that do not directly harm the copyright holders. On the other hand, the Meta case ruling imposes a high evidentiary threshold on plaintiffs and obligates them to present solid evidence of market dilution. Rulings in future lawsuits are likely to shape the boundaries of American, and perhaps even international, copyright law.

Recommendations for Companies in Israel

To date, there is no binding ruling in Israel on this hot topic. However, at the end of 2022, the Israeli Ministry of Justice published a position in principle, whereby the use of copyrighted content for the purpose of machine learning is protected, in most instances, under the permitted fair use arrangements under copyright law, and therefore does not constitute copyright infringement. This position reflects a desire to spur innovation and maximize the competitiveness in the Israeli market for ML and content creation.

Notwithstanding the emerging trends, and due to the current legal uncertainty in this regard, we recommend that companies refrain from relying on the fair use defense when it comes to using copyrighted works to train AI models. We also recommend that companies seek individual, practical advice about the various considerations deriving from case law and presented in government position statements, including examining the materials in their possession to ascertain whether they are legally purchased and controlled or originate from pirated sources.

Recommended practical actions under the current circumstances:

• Isolate legitimate databases from unregulated (pirated) content and ensure you do not use pirated content for machine learning.
• Formulate and act in conformity with straightforward policy documents regulating the use of copyrighted works for machine learning.
• Document the training process, including the source of the copyrighted materials and how you use them, and assess the legal, strategic, and economic risks and considerations.
• Analyze in advance whether use of copyrighted works could dilute the market for each copyrighted work you use and document the findings of your analysis.

***

Dr. Avishay Klein is a partner and heads our firm’s Privacy, Data Protection, Cyber and AI Department.

Dr. Ran Karmi is an associate in our firm’s Antitrust and Competition Department.

Barnea Jaffa Lande’s Privacy, Data Protection, Cybersecurity and AI Department is one of the most prominent and leading practices in Israel and provides comprehensive, innovative legal services to technology companies, institutional entities, and corporations from various sectors in Israel and abroad.