AI copyright news, graded by source

Microsoft asks the court to deny the authors summary judgment, arguing OpenAI's LibGen and Books3 downloads were for transformative training and Microsoft did no downloading

In a 54-page brief filed Oct. 23, 2026 (public redacted version at Dkt. 2173; the full brief is under seal at Dkt. 2179), Microsoft opposes the class authors' motion for partial summary judgment. It argues that training a large language model is fair use as a matter of law, calling an LLM "not a book machine. It is an everything machine." On the authors' lead claim that OpenAI's downloads from LibGen and of the Books3 dataset infringed regardless of training, Microsoft says the purpose of the downloading was to tokenize the books for training, unlike Napster users or the Internet Archive's lending, so it was transformative. It says it did not do the downloading itself and cannot be directly or vicariously liable for it, and, citing OpenAI's own fact statement, that OpenAI's use of shadow-library data in training was limited to GPT-3 and GPT-3.5, before a later Microsoft contract right to demand retraining. It also argues the authors show no lost sales, that their market-dilution theory is speculative, and that there is no cognizable market for training licenses. These are Microsoft's arguments; the court has not ruled.