Dialogue from more than 139,000 movies and TV shows has been used without permission to train generative-AI tools, including by companies such as Apple and Anthropic (of Claude), a new investigation in The Atlantic reveals