Every State Bar's Stance on Lawyers Using AI (2026 Guide)

If you are a lawyer wondering whether you are allowed to use ChatGPT for client work, the answer from every bar that has looked at it is the same: yes, with conditions. No bar has banned it.

What the bars disagree on is narrower, and it is the part worth your attention. They split on whether you have to tell your client you used AI.

Here is where things stand in 2026, and what you actually need to do.

Who has issued guidance

The American Bar Association set the national baseline with Formal Opinion 512 in July 2024. It does not create new rules. It maps the existing Model Rules onto AI and names the duties that apply: competence, confidentiality, communication, candor, fees, and supervision.

Since then, 11 states plus D.C. have issued formal ethics opinions specifically on lawyers using AI. The New York City Bar and the California State Bar have added their own guidance. Counting task force reports and FAQs from states like New Jersey, New York, Illinois, Michigan, Minnesota, and Arizona, more than two dozen jurisdictions have now weighed in.

If your state is not on the list, you are not off the hook. The underlying Rules of Professional Conduct still apply, and most bars treat ABA Opinion 512 as persuasive in the meantime.

What they all agree on

Read enough of these and the same four points come up every time.

You own the output. AI is a starting point, not a source. You have to verify everything it produces, especially case citations. The cautionary tale every opinion cites is Mata v. Avianca, where lawyers were sanctioned for filing a brief full of cases ChatGPT invented. If your name is on the filing, the mistake is yours.

Protect client confidentiality. Before you put client information into a tool, you need to understand what it does with that information. The recurring worry is self-learning models that store your inputs and surface them to someone else later. Check the data retention, data sharing, and training policies first.

Bill honestly. This one trips people up. If AI saves you three hours, you cannot bill the client for three hours you did not work. You also cannot bill clients for the time you spend learning the tool. You can bill for the actual time you spend using and reviewing it on their matter, and you can pass through the tool's per-use cost.

Supervise it like a junior assistant. Most opinions treat AI as the equivalent of a nonlawyer assistant under the supervision rules. You are responsible for its work the same way you are responsible for a paralegal's.

Where they split: telling the client

This is the part that actually varies by state.

Some bars lean toward disclosure or consent. California recommends informing clients when AI is used in their representation. Florida's Opinion 24-1 advises getting informed consent before using a third-party tool that would expose confidential information, and it requires any client-facing chatbot to identify itself as an AI and not a lawyer.

Others decline to make it a rule. Texas Opinion 705 calls for thoughtful analysis but stops short of requiring informed consent in every case. The New York City Bar takes a similar line and says routine, embedded tools like Microsoft Word, Westlaw, and Lexis do not require any disclosure at all.

So the practical answer depends on two things: where you practice, and how the tool handles your client's data. A closed tool that does not send data to a third party raises fewer issues than an open consumer chatbot.

The guidance also keeps moving. California replaced its 2023 guidance in May 2026 to cover agentic AI, meaning tools that act on their own to plan steps and complete tasks. The bar's position is that this autonomy does not relieve you of the duty to exercise your own judgment, so an agent cannot make legal decisions or file anything without your review. Expect more states to address this next.

What to actually do

You do not need to read all fifteen opinions. You need a short routine.

1. Check whether your state has its own opinion. If it does, that controls. If not, use ABA 512 as your baseline.
2. Before entering client data into any tool, confirm its retention and training policies. Prefer tools that do not train on your inputs.
3. Verify every citation and factual claim before it leaves your desk.
4. Bill for time actually spent, not time saved.
5. If your state leans toward disclosure, or the tool is an open consumer product handling confidential facts, tell the client and get consent.

We keep plain-English summaries of these opinions, with the official text and source for each, in our ethics opinions library. It is free, and we update it as new opinions are adopted.

This article is for general information and is not legal advice. Advisory ethics opinions are not binding, and the rules in your jurisdiction control. Confirm the current rules with your bar before acting.