West Virginia AI Ethics Rules for Lawyers

The official instruments

The strictest consent rule in the country

West Virginia is the national outlier, and on the single issue that matters most to clients, consent. Legal Ethics Opinion 24-01, a formal opinion of the West Virginia Lawyer Disciplinary Board issued in June 2024, holds that client consent to generative-AI use must be informed and confirmed in writing. No other U.S. instrument this tracker has verified goes that far. That one sentence reorders where West Virginia sits on every comparison chart: it is not the state with the most detailed guidance or the first to act, but it is the state that demands the most before a lawyer may use these tools on a client’s matter.

The distance between West Virginia and the rest is best seen by lining up the three postures the country has actually produced. At one end are the no-disclosure-duty jurisdictions: New Jersey’s Preliminary Guidelines state there is no blanket duty to tell clients about AI use, and the Illinois Supreme Court’s policy declines to require AI disclosure in pleadings at all. In the middle sits the national mainstream, which conditions consent on exposure: states like Texas and Oregon require a lawyer to consider or obtain client consent before confidential information is fed to a tool, tying the duty to the confidentiality analysis rather than to AI use as such. West Virginia occupies a third position by itself. It does not merely ask lawyers to consider consent when confidences are at risk, and it does not settle for oral agreement; it requires consent to generative-AI use that is both informed and confirmed in writing. The trigger is broader and the formality is higher than anywhere else. For a multi-state firm, that means West Virginia cannot be folded into a general consent policy built for the mainstream; it needs its own written-confirmation step, and the state comparison table exists precisely to keep this outlier from being missed.

AI as a secondary source, and what the Board demands around it

The written-consent rule does not stand alone; it sits inside an opinion that is unusually skeptical of the technology itself. LEO 24-01 frames generative AI as at best a secondary source, a tool that may supplement but never replace a lawyer’s own reasoning. That framing is doing more than setting a tone. It assigns the work product to the lawyer’s judgment and treats the model’s output as raw material that has not yet been thought about. Diligence, the opinion holds, requires human oversight and intervention to correct AI mistakes, so the duty is not satisfied by a passive review that rubber-stamps whatever the tool produced; it requires the lawyer to catch and fix errors actively. A West Virginia lawyer who lets AI output stand as the analysis, rather than as input to the analysis, has not met the standard the Board describes.

Around that core, the opinion imposes the duties the national consensus shares, but in West Virginia’s demanding register. Lawyers must vet a tool’s terms of use and privacy policies before using it, which is the confidentiality-and-data-handling check every serious instrument requires, framed here as a precondition rather than a suggestion. Lawyers must supervise AI the way they supervise nonlawyer assistance, locating responsibility for the tool’s work squarely with the supervising lawyer. And lawyers must monitor tools for errors and bias on an ongoing basis, not merely at adoption, recognizing that a tool’s behavior can drift and that bias is a live risk rather than a theoretical one. Taken together with the written-consent requirement and the secondary-source framing, the composite West Virginia posture is the most cautious in the country: the lawyer must get written consent up front, treat the output as unproven until verified, keep a human hand on every result, and watch the tool over time.

What West Virginia lawyers should do now

The build follows directly from the outlier rule. Put a written-consent step at the front of the engagement, because in West Virginia consent to generative-AI use must be informed and confirmed in writing, and an oral understanding or a silent assumption will not do; engagement-letter language that explains the use and captures the client’s written agreement is the cleanest way to satisfy it. Maintain an approved-tools register that records each tool’s terms of use and privacy policy, so the vetting duty is documented rather than assumed, and prefer tools whose data handling can be explained to a client clearly enough to make consent genuinely informed. Enforce a verification workflow that treats every AI output as a secondary source requiring independent confirmation, with a human reviewer responsible for catching and correcting mistakes. Supervise firm-wide use through a written, acknowledged policy, and schedule periodic monitoring for errors and bias rather than treating adoption as the end of the inquiry. The policy template implements each of these elements, and because West Virginia’s consent rule is the strictest in the country, the written-confirmation step is the one a firm operating in multiple states is most likely to omit by default; the tracker and its changelog will move this entry the week anything changes it.

One layer LEO 24-01 does not occupy is the courtroom. West Virginia has no statewide court rule on generative AI that this tracker has verified, so the obligations that exist at the filing stage are judge-by-judge standing orders rather than a uniform state command. That means a West Virginia lawyer faces two distinct consent and disclosure questions: the bar’s written-consent rule, which runs to the client, and any individual judge’s standing order, which runs to the court and may demand its own certification or disclosure. Confirm the assigned judge’s standing orders before every filing, because a chambers-level requirement can exist even though the state has issued no court-wide rule. The court orders guide covers that per-filing check and the kinds of orders to watch for.