Models including Anthropic’s Claude 3.7 Sonnet and OpenAI’s GPT-4o explicitly refuse to remove watermarks; Claude calls removing a watermark from an image “unethical and potentially illegal.”

Removing a watermark without the original owner’s consent is considered illegal under U.S. copyright law (according to law firms like this one) outside of rare exceptions.

Is training your AI with copyrighted data unethical and potentially illegal too? Or do these AI companies get a free pass?