The problem with "AI" in legal --- all ethical problems aside --- is that legal is heavily reliant on language in addition to logic. Determining the "correct" use of legal language is a difficult and ill-posed problem with multiple viable solutions and no means to safely choose between them without authoritative case law coming to the rescue.
Verifying legal solutions is a problem with a high human verification burden, so producing mass "AI" solutions mostly increases human workload, instead of reducing it.