Text AI watermarks will always be trivial to remove

I presume all the Foundation AI companies have been doing this for years, to avoid training models on their own generated data?