They also initially took content from libgen, which is a fair bit less legal. Personally, I have mixed feelings about all of this. On the one hand, I don't like some shitty for-profit AI company making money from the collective works of civilisation. On the other hand, I think copyright protects works for far too long anyway and most should be in the commons already. Mind you, I would be more sympathetic if Anthropic et al. were doing all this for research purposes instead of capitalism. Maybe that would be a better copyright reform, in that it expires much more quickly than the current laws (say 10 years) but restricts third parties making a profit for a longer period. Likely that would be complex to design and enforce, however.