Meta’s LLaMA AI Accused of Repeating Books Verbatim, Raising Legal Risks

Stanford law professor Mark Lemley found that Meta’s LLaMA AI can reproduce entire copyrighted book passages verbatim, unlike other models tested. His team’s research showed that LLaMA memorized content from works like Harry Potter, 1984, and The Great Gatsby, raising concerns that it functions more like a piracy tool than a generative AI. The findings could significantly damage Meta’s legal defense, especially since LLaMA was trained on the controversial Books3 dataset obtained via torrent.