Sony and Warner accuse Anthropic of training Claude on tens of thousands of pirated works. Should the model be retrained from scratch?

Sony Music Publishing and Warner Chappell allege that Anthropic used mass torrenting, scraping, and downloading to train Claude. Anthropic disputes the claims and says it will defend itself.
A fine could simply become the cost of doing business. But forcing a company to discard or retrain a model could reshape the entire AI industry.
What would actually be fair here: licensing fees, damages, or retraining from scratch?