Meta is currently entangled in a copyright lawsuit that alleges the company utilized copyrighted materials without permission to train its artificial intelligence (AI) models. The case has been brought forward by a coalition of complainants, including several well-known authors. The central claim asserts that Meta employed unauthorized e-books and articles to develop earlier versions of its Llama AI models, constituting a breach of copyright regulations. Furthermore, the suit implicates company CEO Mark Zuckerberg for purportedly permitting the Llama AI team to access a controversial link aggregator for obtaining these copyrighted materials.
This information emerged from two separate filings submitted to the U.S. District Court for the Northern District of California on Wednesday. The documents, filed by authors including Sarah Silverman and Ta-Nehisi Coates, refer to testimony from Meta in late 2024 that reportedly revealed Zuckerberg authorized the use of a dataset known as LibGen for training its Llama AI models.
LibGen, or Library Genesis, is a file-sharing platform that provides free access to a range of academic and popular content. It is frequently labeled as a pirate library, due to its provision of copyrighted materials that are either paywalled or not digitized. The platform has undergone numerous legal challenges and has been subjected to shutdown orders over the years.
The lawsuit asserts that Meta knowingly used content from the LibGen dataset, aware that it contained pirated material and violated copyright laws. The documents also allege that a memo addressed to senior personnel in Meta’s AI division indicated that after an “escalation to MZ,” an acronym for the CEO’s name, the AI team was granted permission to utilize LibGen.
Moreover, the memo alerted executives about potential public backlash concerning the use of “a dataset we know to be pirated such as LibGen,” particularly how this could negatively impact Meta’s discussions with regulators. The filings contend that the company took steps to remove copyright metadata and information from the dataset to hide its infringement.
According to the documents, Nikolay Bashlykov, a research engineer at Meta, was accused of erasing copyright information from the LibGen dataset. To further obfuscate the origins of the material, Meta’s developers allegedly incorporated “supervised samples” of data during the fine-tuning process of Llama, ensuring that the model produced less problematic answers when queried about its training data sources.
Additionally, the complaint asserts that merely accessing LibGen constituted a form of copyright infringement. The filings suggest that Meta downloaded and shared the LibGen dataset through torrenting, a process that entails both downloading and uploading content, which can be interpreted as distributing copyrighted materials—another violation of copyright laws.
The filings critiqued Meta’s decision to circumvent lawful avenues for acquiring these works, stating, “Had Meta bought Plaintiffs’ works in a bookstore or borrowed them from a library and trained its Llama models on them without a license, it would have committed copyright infringement. Meta’s choice to engage in and promote an illegal torrenting network illustrates a violation of the California Comprehensive Computer Data Access and Fraud Act (CDAFA) and serves as evidence of copyright infringement.”
The copyright lawsuit remains active, with a ruling forthcoming. Meta has yet to present its defense, expected to invoke fair use principles. The court will need to determine whether the generative nature of the AI model is transformative enough to validate such an argument.
