Can Lawyers Use ChatGPT? The 2026 Answer

The short answer has not changed; the conditions have multiplied

Since California issued the first state guidance in November 2023, every U.S. ethics authority to address the question has reached the same conclusion: lawyers may use generative AI. What has changed between 2023 and 2026 is the density of conditions around that yes. Sixteen formal ethics opinions, ten official guidance documents, three reports, and two court-system policies now apply to how lawyers use these tools, and as of June 1, 2026, New York lawyers practice under the first binding statewide court rule on the subject. The tracker maintains the full, honestly labeled list.

The conditions sort into five questions you can actually operationalize.

1. What are you putting in?

The confidentiality analysis is the same everywhere even when the consent threshold differs. A consumer chatbot that trains on inputs is a third party; information relating to a representation that goes in may not stay private. So: know the tool’s terms (training, retention, access), prefer enterprise tiers with training disabled, anonymize where the task allows, and obtain informed consent where meaningful risk remains. Florida’s Opinion 24-1 recommends consent before feeding confidential information to third-party AI; Oregon distinguishes open from closed models and requires consent for the open ones; West Virginia’s LEO 24-01 requires informed consent confirmed in writing, the strictest position in the country. New Mexico adds a practical point others skip: anonymization means more than deleting names, because fact patterns identify clients too. Full breakdown in the confidentiality guide.

2. What are you doing with what comes out?

Treat every output as a first draft from a bright, unsupervised assistant who occasionally invents authority with total confidence. The verification duty appears in every instrument on the tracker: review outputs before use, and independently verify every citation, quotation, and statement of law in a traditional research database before filing. New Mexico’s opinion makes explicit what should be obvious: asking the same AI whether its citations are real is not verification. The sanctions caselaw, from Mata v. Avianca forward, is a catalog of lawyers who skipped this step; see the candor and sanctions guide.

3. What are you billing?

Hourly billing and AI efficiency interact one way: the client gets the gains. Bill time actually spent prompting, reviewing, and correcting. Do not bill the research time the tool eliminated, and do not bill clients for learning the tool absent a specific agreement. Pass through AI costs only where reasonable, matter-specific, and agreed in advance. Kentucky, North Carolina (which puts it bluntly: no charging three hours for a one-hour AI-assisted task), Texas, and the ABA all converge here, with Virginia’s LEO 1901 adding that flat and value-based fee arrangements may legitimately capture efficiency gains. Details in the billing guide.

4. Who else in your firm is using it?

You are responsible for the answer. Rules 5.1 and 5.3 extend your duties to subordinate lawyers, staff, and vendors using AI on firm matters. Arizona’s best-practices guidance and Missouri’s Informal Opinion 2024-11 both land on the same mechanism: written firm policies, training, and acknowledgment. If your firm has no AI policy, the supervision duty is currently unmanaged; the policy guide and free template close that gap.

5. Where are you filing?

Ethics opinions are only half the map. The other half is court-by-court: standing orders and local rules that require disclosure of AI use, certification of citation verification, or in a handful of courtrooms prohibit AI-drafted filings outright. There is no national pattern; New York’s Part 161 declines to require disclosure system-wide while individual judges elsewhere require it per filing. Check the judge, every time. The court orders guide explains the layer and points to the free trackers that catalog it.

The honest bottom line

The question in 2026 is not whether you can use ChatGPT; it is whether your use survives five specific audits: inputs, outputs, bills, supervision, and the forum. A lawyer who can pass all five is not just compliant but better positioned than the majority of practitioners who still treat AI policy as someone else’s problem.