California Wants Every Lawyer to Prove They Checked AI's Work. Most Firms Can't.
A lawyer cites a case that doesn't exist. A judge catches it. The lawyer says an AI tool wrote it and they didn't check closely enough. That scenario has played out in courtrooms across the country enough times that California's State Bar decided a reminder wasn't going to cut it.
On March 13, 2026, the Committee on Professional Responsibility and Conduct (COPRAC) approved proposed amendments to six Rules of Professional Conduct, all aimed at how lawyers use AI. The rulemaking traces back to an August 22, 2025 letter from the California Supreme Court asking the Bar to build on its 2023 generative AI guidance, and to go further into "agentic AI," tools that plan and execute tasks with little direct human input. Public comment on the first draft closed May 4, and the committee approved a revised version on June 12 with a second comment round that ran through August 6.
What the proposed rule actually says
The changes land in the official Comments to the rules, not the black-letter text, but Comments are exactly what a State Bar court points to when deciding if conduct was sanctionable. Six rules are touched:
- Rule 1.1 (Competence): a new comment would require a lawyer to "independently review, verify, and exercise professional judgment" regarding any output generated by technology, AI included.
- Rule 1.4 (Communication with Clients): when AI materially affects the representation, the lawyer has to tell the client enough to make an informed decision, and keep telling them as the matter goes on.
- Rule 1.6 (Confidentiality): feeding client information into an AI tool counts as "revealing" it if doing so creates a material risk the tool (or its operator) uses that information inconsistently with the duty of confidentiality.
- Rule 3.3 (Candor to the Tribunal): a lawyer must confirm that cited authorities actually exist and say what the lawyer claims they say, whether a human or an AI tool drafted the citation.
- Rules 5.1 and 5.3 (Supervision): managing lawyers need written policies on AI use, and have to train and supervise nonlawyer staff on the ethics of using it.
None of this bans AI tools. It assumes you're using them and asks you to prove you checked the output, the same way you'd be expected to check a paralegal's draft.
Why "verify" is the hard part
Most small firms that use AI today do it the same way: copy a fact pattern into a chat tool, read the output, and either use it or don't. There's rarely a record of what was reviewed, what was changed, or when. If a client, opposing counsel, or a bar investigator ever asks "how do you know you checked this," the honest answer in a lot of offices is: we don't have anything that shows it.
That's not a technology problem. It's a process problem, and it existed before AI showed up. Firms already lose track of who reviewed a document before it went out. AI just raises the stakes, because the proposed Rule 3.3 comment puts the citation-checking duty on the record explicitly.
This isn't just a California problem
California is the state furthest along on specifics, but it's not alone. More than a dozen state bars and the District of Columbia have already issued formal ethics opinions on AI use, and Florida's Ethics Opinion 24-1 separately requires a client-facing AI chatbot to disclose that it's AI. North Carolina and most other states haven't gone as far as a Comment-level verification duty yet, but the direction is consistent: if you use AI in client work, your state bar increasingly expects you to be able to show you reviewed it, not just assert that you did.
What to put in place now
You don't need to wait for your state to finalize a rule to fix the underlying gap. A few things make the proof problem go away:
- Keep a dated log of what an AI tool drafted and what a human changed before it went out, even a simple one.
- Separate "draft" from "sent." Nothing an AI tool produces (a letter, a filing, a client update) should leave the building without a named person signing off.
- Write down, in one page, what AI tools your firm uses, for what tasks, and who's responsible for checking the output. That's most of what Rule 5.1 is asking for.
This is the same model we build around for every firm we work with: an agent drafts the document, flags the citations, or prepares the client update, and a person at the firm reviews and approves before anything goes out. That sequence is also the audit trail a bar investigator would want to see. See how the review step works, and check our legal page for how we handle intake and documents for law firms specifically.
Sources: State Bar of California, Proposed Amendments to the Rules of Professional Conduct Related to Artificial Intelligence and LawNext, "California Bar Proposes Rule Requiring Lawyers to Verify Every AI Output".