Anthropic trained Claude on illegally obtained Stanford professors' books

Discussion overview

The thread spent less time on Stanford professors specifically than on the legal and moral split between acquiring books illegally and using books to train a model. Several commenters stressed that the reported ruling treated training on lawfully obtained books as transformative fair use; in their view, Anthropic’s expensive mistake was sourcing material through LibGen rather than buying and scanning it. That distinction mattered because it suggests a route for large-scale model training that avoids treating every learned idea as licensed property.

Others called that distinction evasive. They argued that pirating works expressly to build a commercial model cannot be neatly separated from the resulting training process, especially when models can operationalize an author’s accumulated expertise at enormous scale. The most forceful criticism was that AI firms have extracted broadly from creators while reserving tight control over their own models, making objections to model distillation look hypocritical. A smaller group welcomed the erosion of academic and publishing gatekeeping, but opponents replied that broad access does not justify stripping authors of compensation. Commenters also rejected the article’s original framing as legally imprecise: the alleged illegality concerned acquisition, while the fair-use finding on training remains a crucial, contested precedent rather than a blanket moral acquittal.

Points of disagreement

The core fight is whether lawful training can be meaningfully separated from pirated acquisition. One side says the judge correctly recognized model training as transformative learning, so the remedy should target LibGen sourcing and leave properly acquired corpora usable. The other says a commercial AI system built from deliberately stolen books is inseparable from that theft, and fair use lets companies capture creators’ expertise without a fair bargain.