Meta Wins AI Copyright Case, Judge Issues Future Cautions on Risks
In a landmark decision, U.S. District Judge Vince Chhabria has granted summary judgment in favour of Meta, the tech giant behind Facebook, in a lawsuit brought forth by 13 authors, including comedian Sarah Silverman and writer Ta-Nehisi Coates, who accused the company of copyright infringement. The authors claimed that Meta used their works to train its AI systems without obtaining prior consent.
The court's ruling shielded Meta from copyright liability in this specific case, marking a significant step forward in the evolving legal landscape around using copyrighted material for AI training. The decision leans towards considering such use as fair use under U.S. copyright law, but important nuances remain.
Judge Chhabria's decision follows a series of federal court decisions, including from the Northern District of California and San Francisco, which have ruled that training AI models on copyrighted works is a transformative use considered fair use. These rulings emphasize that the purpose of training AI, such as developing large language models, is transformative and does not infringe copyright when done on lawfully acquired materials without distributing copies.
However, using pirated or illegally downloaded materials is explicitly not fair use, regardless of intent. If the AI-generated outputs themselves directly copy or infringe copyrighted content, the legal situation may differ, potentially leading to infringement claims.
Despite some preliminary rulings supporting fair use, many lawsuits remain pending, and the definitive legal consensus has not yet been reached. Notable cases like Kadrey vs Meta and Bartz vs Anthropic are still ongoing, highlighting the uncertainty in the space.
The U.S. Copyright Office currently holds that generative AI output is generally not copyrightable due to the lack of human authorship, but users of AI-generated content still face potential legal liabilities, especially if the output contains large segments of copyrighted text or images of trademarked or copyrighted characters.
Industry actors are proactively adjusting practices to comply with evolving legal standards while advocating for clearer rules around AI training data. Meta and OpenAI (and also Anthropic) have been involved in cases testing these boundaries. For example, Anthropic was ruled not to have violated copyright law in using copyrighted books to train its large language models.
Many AI developers emphasize transparency in data sourcing and implement practices to respect copyright, such as avoiding pirated content and destroying source copies after digitization where applicable. Some companies advocate for or await legal and market-driven frameworks to clarify rights and obligations around copyrighted materials in AI training.
While the New York Times is a frequent plaintiff in lawsuits about unauthorized use of its copyrighted materials in AI training data, no direct ruling has yet conclusively resolved its claims. The Times objects primarily to scraping and using its proprietary journalistic content without consent, which is a common practice by many AI companies using internet-scraped data. The current rulings provide a framework but leave open many questions specific to individual content owners like the Times.
Meta's spokesperson welcomed the court's decision, supporting the principle of fair use as a vital legal framework for building transformative AI technology. OpenAI maintains that its activities fall within the bounds of fair use and emphasizes the importance of advancing AI technology. The specific case involved the use of the authors' works to train AI models, not the use of these models to generate content.
Judge Chhabria's remarks offered a measure of reassurance to creative professionals who argue that such practices infringe on their intellectual property rights. However, the legal landscape around AI and copyright continues to evolve, with ongoing litigation and the need for clearer guidelines to ensure fair use and protect intellectual property rights.
The court's decision to consider the use of copyrighted material for AI training as a transformative use supports the principle of fair use, promoting the development of AI technology. However, the use of pirated or illegally downloaded materials is still not considered fair use, and legal nuances remain in this evolving landscape, as notable cases like Kadrey vs Meta and Bartz vs Anthropic continue to unfold.
