How would a DMCA takedown against a machine learning model work? Is it possible to build it in a way where you can easily reverse the effects of any given stimu

How would a DMCA takedown against a machine learning model work? Is it possible to build it in a way where you can easily reverse the effects of any given stimulus, or would you need to start completely from scratch? This seems like a problem similar to explainable AI.