This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Meta and OpenAI have been accused in court filings of using the pirated 'shadow library' LibGen to train their generative AI models. Internal Meta communications suggest that the decision to use this pirated material was escalated to CEO Mark Zuckerberg, with employees discussing methods to mask the dataset's origins and avoid legal scrutiny. The use of these datasets has sparked significant backlash from authors and publishers, leading to class-action lawsuits and public protests regarding copyright infringement and the unauthorized exploitation of creative labor.
Court records reveal that Meta employees allegedly discussed pirating books to train LLaMA 3, citing cost and speed concerns with licensing. Internal messages suggest Meta accessed LibGen, a repository of over 7.5 million pirated books, with apparent approval from Mark Zuckerberg. Employees allegedly took steps to obscure the dataset’s origins. OpenAI has also been implicated in using LibGen.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems capable of creating economic or cultural value, including through reproduction of human innovation or creativity (e.g., art, music, writing, coding, invention), destabilizing economic and social systems that rely on human effort. The ubiquity of AI-generated content may lead to reduced appreciation for human skills, disruption of creative and knowledge-based industries, and homogenization of cultural experiences.
Also classified as
Meta employees actively discussed removing copyright markers and hiding the use of LibGen to avoid regulatory and legal scrutiny.
The unauthorized use of millions of copyrighted books to train AI models devalues the creative labor of authors and threatens to disrupt the creative writing industry.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Pre-deployment
Occurring before the AI is deployed
The decision to download and train the AI models on pirated LibGen data was intentionally made by humans at Meta and OpenAI during the pre-deployment training phase.
Added categories peak at 3: Substantial (Epistemic Harm, Inferred)
Meta and OpenAI allegedly used pirated material from the shadow library LibGen to train their generative AI models. While this presents significant intellectual property and economic challenges for the creative sector, it does not pose a direct threat to national security, critical infrastructure, or sovereignty.
The report describes the AI systems as generative-AI chatbots and assistants, such as Meta AI, which are designed to answer queries and generate creative content to augment human capabilities.
Risk Level 3: The systems involved are generative AI chatbots and assistants, which fall under the Limited Risk category requiring transparency obligations.
The models trained (Llama 3 and OpenAI models) are primarily deployed as general-purpose chatbots and writing assistants to generate text and answer user queries.
The reports provide detailed, unsealed court documents and internal communications from Meta detailing the decision-making process, the specific dataset used (LibGen), and the reactions of the affected authors. The facts are well-documented across multiple reputable journalistic sources.
The incident could be viewed purely as a standard copyright dispute rather than an AI safety risk, though the scale and nature of LLM training make it a distinct AI governance issue.
Technology
High Data Demands for LLM Training
Llama 3 required massive high-quality text to compete with rival models.
BitTorrent Protocol Distribution
Torrenting LibGen automatically uploaded copyrighted files to other peers.
Data Inputs
Use of Pirated LibGen Dataset
Meta downloaded 7.5 million books from a notorious shadow library.
Removal of Copyright Metadata
Employees stripped ISBNs and copyright notices to mask data origin.
Human Factors
Squeamishness Over Torrenting
Engineers felt downloading pirated data on corporate laptops was wrong.
Apathy Toward Creator Rights
Prioritizing model performance over fair compensation for authors.
Process and Methods
Active Obfuscation Tactics
Stripping copyright markers and fine-tuning models to refuse quoting books.
Flawed Fair Use Strategy
Avoiding licensing to maintain a broad 'fair use' legal defense.
Management
Prioritizing Speed Over Ethics
Executives demanded books ASAP, rejecting slow legal licensing paths.
Executive Approval of Piracy
CEO Mark Zuckerberg approved the use of the pirated LibGen dataset.
Rejection of Licensing Costs
Management deemed licensing high-quality data to be too expensive.
Regulatory Environment
Unclear Fair Use Boundaries
Ambiguity in copyright law regarding AI training encouraged risky choices.
Failure to Shut Down LibGen
Courts ordered fines but failed to stop the online shadow library.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
