Maybe both are ok. Maybe neither. Maybe one. The point is this does not follow >If training on copyrighted data without authors consent is ok then distilling i

Maybe both are ok. Maybe neither. Maybe one. The point is this does not follow >If training on copyrighted data without authors consent is ok then distilling is ok as well. Because building a model from distillation and building a model from raw data are not the same. You have to evaluate them independently. And legally its different as well. IP vs ToS (civil).