Isn't there sufficient evidence to conclude that OpenAI is training on things they know they probably shouldn't be training on? Like the full text of novels tha

Isn't there sufficient evidence to conclude that OpenAI is training on things they know they probably shouldn't be training on? Like the full text of novels that are not publicly available on the internet? I don't think it's that unreasonable to extrapolate that they would blur the lines of legal agreements to get access to private user data they shouldn't have.