What Claude Can't Do for Lawyers (And Why That Matters)

Every piece I've written about using Claude in a legal practice comes with caveats, and I want to put them all in one place instead of scattering them across ten articles. If you only read one thing about AI and legal work, I'd rather it be the honest limits than another list of things it's good at.

It doesn't know your jurisdiction's current law

Claude's knowledge has a training cutoff, and even within that window, it cannot reliably tell you that a specific statute, rule, or case is still good law today. Legislatures amend, courts overrule, rules of procedure change. Anything that matters to a filing needs to be checked against a current, authoritative source — not because Claude is unusually unreliable, but because no static training process can track a live legal system in real time.

It doesn't hold privilege the way you think it might

Depending on how you access Claude and your firm's data agreements, information you paste in may or may not be used to improve the model, and may or may not be adequately protected for privileged material. Before pasting client information into any AI tool, understand your specific product's data handling terms, your jurisdiction's ethics rules on client confidentiality and AI, and your firm's own policy. This is not a place to guess.

It can sound certain while being wrong

This is the one that catches experienced lawyers off guard, because it doesn't match their intuition about when to be skeptical. A junior associate who's unsure usually sounds unsure. Claude, by default, often states a wrong answer with the same confident tone as a right one. The fix isn't distrust of everything it says — it's building the habit of verifying specifically the claims that matter (citations, deadlines, jurisdiction-specific rules) rather than trusting tone as a signal of accuracy.

It doesn't replace judgment on a close call

Whether to take a marginal case, how hard to push in a negotiation, when a client relationship needs a phone call instead of an email — these are judgment calls built on pattern-matching from actual experience, including experience of being wrong before. Claude can lay out considerations, but the decision, and the responsibility for it, stays entirely with the lawyer.

The honest summary

• Use it for structure, drafting speed, and getting oriented — not for facts that need to be current or citations that need to be real.
• Check your data handling and confidentiality obligations before pasting in anything client-related.
• Verify tone-independent of confidence — a confident answer and a correct answer are not the same thing.
• Keep judgment calls where they belong: with you.

The lawyers who get burned by AI aren't the ones who use it. They're the ones who stop checking, because it started sounding right often enough that checking felt unnecessary. Don't let it feel unnecessary.