The recent decision by the United States District Court for the Northern District of California in Bartz v. Anthropic PBC is helping shape AI copyright law in the United States. In the decision, the Court evaluated Anthropic’s use of copyrighted material to train its AI model, Claude.

Under U.S. copyright law, the “fair use” doctrine allows for “transformative” uses of a work protected by copyright. The Court found that training a large language model (LLM) on copyrighted books was “exceedingly transformative.” It compared the training process to a human author being influenced by works they have read, thus qualifying as fair use. However, the Court distinguished between training the LLM on legally obtained works and works obtained through piracy, finding the former to be acceptable fair use and the latter a violation of copyright.

How a similar case would be handled in Canada is unclear. Under Canadian copyright law, the doctrine of “fair dealing” allows for the use of a copyrighted work without requiring permission of the copyright owner. The first step in assessing fair dealing is to determine whether the use fits into one of the statutory categories set out in the Copyright Act: research, private study, education, parody, or satire. To date, Canadian courts have not determined whether use of a copyrighted work for training a LLM qualifies as one of these categories.

A copy of the full decision is available here.