I wonder if it would be feasible to train a neural network on a large corpus of test cases where legality/illegality have already been defined, then release thi

I wonder if it would be feasible to train a neural network on a large corpus of test cases where legality/illegality have already been defined, then release this neural network to the public. Sure, the process will be opaque, but it will be consistent between different users, and answers can be sought almost immediately with little to no costs. In many cases, an immediate straight up no is more valuable than a maybe suspended over months or years.