The Catch With Legal AI

AI is already doing real work inside law firms. Document review, contract analysis, first-draft NDAs. Thomson Reuters clocks about four hours a week back per lawyer. That part is settled, and it's the reason 73% of legal professionals say they plan to use it.

But there's a catch in legal AI that doesn't get said enough, and it flips who the tool actually helps.

To use AI safely on legal work, you have to already know enough to catch it when it's wrong. AI output reads as authoritative and is sometimes invented. It cites cases that don't exist. It applies the wrong jurisdiction. A fifteen-year litigator spots it in a second. A first-year associate might not. A non-lawyer almost certainly won't. So the tool helps the people who least need it and quietly endangers the ones who most do.

The New York State Bar Association put it plainly in 2025: AI output has to be verified by someone with the legal knowledge to judge it. If you can't do the work yourself, you can't safely supervise the machine doing it.

That isn't an argument against legal AI. It's an argument for the kind you can check. An answer you can trace to a real source, with the citation attached, is one a lawyer verifies in seconds. An answer that arrives as a confident black box is one they have to redo from scratch, which defeats the point. And it only stays inside the duty of confidentiality if the client's data never left the firm to get the answer.

We pulled the full 2026 data on AI and the legal profession, what it does, where it stops, and which practice areas feel it first, into one piece in the Knowledge hub: Will AI Replace Lawyers? What the Data Says in 2026.

Or if the part you care about is private, citable AI that keeps client data inside the firm, here's what that looks like for legal teams.