AI copyright news, graded by source

Meta says any uploading while it torrented book datasets was de minimis, and admits it weighed licensing from publishers before turning to what it calls publicly available data

Meta's answer to the textbook authors' complaint pleads, as its fifth affirmative defense, that to the extent it uploaded portions of copyrighted works while downloading training datasets by BitTorrent, that copying was de minimis and did not produce an identifiable or usable copy of any work. Meta admits it contacted certain publishers and discussed licensing internally, then decided to train some Llama models on what it calls publicly available data. It also admits using its own web crawler, SpiderMate, to gather training data; obtaining data from publicly available sources in 2024; and, beginning with Llama 2, building technology based on some Llama models into WhatsApp, Instagram, Messenger, Facebook and its smart glasses. Other defenses include the three-year statute of limitations, lack of volitional conduct and copyright misuse. The answer also states that Mark Zuckerberg was dismissed from the case on September 3, 2026. These are Meta's own positions; the court has not ruled on them.