This article is three-years old, but it reveals the mindset that some (all?) of the AI-firms have toward the use of copyrighted material for model training.

https://www.businessinsider.com/generative-ai-copyright-meta-google-openai-a16z-microsoft

A problem, from their perspective, is that they need new, human-generated, content to improve the models. Otherwise, they'll eventually be training on mostly AI-generated content and run into the old computer science concept "GIGO" (garbage in, garbage out).

RadioActive wrote:

The news publishers, including the Toronto Star, the Globe and Mail and the Canadian Press, accused the Artificial Intelligence giant of using its copyrighted reporting to train its "Claude" interface without compensation or permission. 

If a court rules in their favour, it could mean billions of dollars for the northern news services. 

Canadian news publishers sue Anthropic over AI copyright infringement