Newly unsealed documents suggest that Anthropic was well aware that destroying books to train its AI would look bad.