As I understand it, there is substantial question as to whether that actually stops them training on your data, it just perhaps changes what derivative processes are applied and used.