OpenAI and Google have suggested tech firms should be able to legally train their models on publicly available data, even if it’s copyrighted.