Meta uses user data and copyrighted works to train AI without consent
Meta reportedly used a pirated book dataset to train its Llama AI model with CEO approval despite internal concerns, and automatically generates AI images of users from their uploaded content. Authors including Sarah Silverman and Ta-Nehisi Coates have sued Meta for training AI on copyrighted works without permission.
What happened
In June 2024, Meta updated its privacy policy to allow use of user information across Meta-related services to train AI models, meaning publicly shared Facebook posts, Instagram photos and captions, and third-party data can be used for training.
Meta CEO Mark Zuckerberg approved the use of LibGen, a dataset providing access to copyrighted works from publishers including Cengage Learning, Macmillan Learning, McGraw Hill, and Pearson Education, to train Llama AI models despite concerns from Meta's AI team that LibGen was a 'data set we know to be pirated.'
Instagram began testing a feature where Meta's AI automatically creates images of users in different scenarios from previously uploaded content and displays them on users' feeds.
Reported impact
Users reported AI-generated images of themselves appearing on Instagram feeds, with questions raised about whether using Meta's AI features grants the company rights to use all user content.
Meta's licensing practices allow the company to create derivative works from user-generated content and sublicense intellectual property rights to third parties without explicit user consent.
Index analysis
Meta's June 2024 privacy policy update grants the company broad rights to use publicly shared user content for AI training without individual consent, extending to posts, photos, captions, and third-party data.
The use of LibGen for Llama training raises questions about whether Meta relied on fair use doctrine to justify training on copyrighted works without publisher permission, despite internal recognition that the dataset contained pirated material.
Governance controls
Meta's terms of service state the company does not claim ownership of user content but grants itself a 'non-exclusive, royalty-free, transferable, sublicensable, and worldwide license' to use, distribute, modify, and create derivative works from user content.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.