Does SB 574 ban lawyers from using AI? What California's new law actually requires

No. SB 574 does not ban California lawyers from using AI, whatever the headlines calling it a crackdown suggest. What it actually does is take duties that already exist under the ethics rules and the sanctions statute and write them straight into the Business and Professions Code and the Code of Civil Procedure. The bill cleared the Legislature on August 31, 2026, and Governor Newsom signed it on September 30. His office listed it among the bills signed that day, and Holland & Knight puts the effective date at January 1, 2027.

We first published this on September 24, before the governor acted, and updated it on October 14. The update corrects one point: the signed text requires disclosure of AI use to the court, which our first version described only as something a lawyer should consider. We could not retrieve the chaptered text from the Legislature's own site, so the section wording below rests on the bill text we read in September and on law firm summaries of the signed version. Check the chaptered text before you quote a subdivision. Here is what the signed bill locks in, and what it leaves exactly where it was.

What the bill actually adds to a lawyer's duties

New Business and Professions Code section 6068.1 folds four obligations into the same statute that already lists an attorney's core duties, the one a disciplinary case cites first. A lawyer using generative AI in representing a client has to keep confidential and nonpublic client information out of a system that doesn't restrict who can see it. A lawyer has to verify the accuracy of anything the tool produces and correct hallucinated content before it goes anywhere near a client or a court. A lawyer has to tell the court when generative AI was used on a document submitted to it, and should consider telling the audience when AI produced public-facing material. And a lawyer may not delegate the practice of law to the tool.

Earlier versions of the bill, and the Senate Judiciary analysis of it, also required lawyers to avoid AI use that produces unlawful discrimination or disparate impact against a protected class. The summaries of the signed version we found, from Holland & Knight and Alston & Bird, do not mention that duty, so we cannot confirm it survived into the chaptered text. Read the final text before you rely on it either way.

None of that is a new idea. Confidentiality, competence, candor and nondiscrimination are already Rules of Professional Conduct duties in California. What 6068.1 does is give a court and the State Bar a second statute to point to, one that names AI specifically instead of making a prosecutor argue from a general competence comment.

The citation rule isn't new, it just moved into the sanctions statute

The part of SB 574 getting the most attention amends Code of Civil Procedure section 128.7, California's version of the federal Rule 11 sanctions statute. The new text says a filed brief, pleading, or motion "shall not contain any citations that the attorney responsible for submitting the pleading has not personally read and verified, including any citation provided by generative artificial intelligence."

Read that again. It doesn't say verify AI citations. It says verify every citation, and then it specifically pulls AI-sourced ones inside that duty so nobody can argue the tool did the reading for them. That is the same duty a lawyer already had under 128.7's existing certification requirement, made explicit for a specific failure mode. Every case in our piece on 2026's disbarment and suspension decisions turned on exactly this: a filed citation nobody had actually read. California just wrote the excuse out of the statute in advance.

The confidentiality test is about access, not the tool's name

Here's where most coverage of this bill gets sloppy. Outlets keep summarizing 6068.1 as "don't use public AI tools," as if the line runs between a free chatbot and some enterprise product with a nicer logo. That's not what the text does. The duty attaches when access to the information is not restricted to people bound by confidentiality obligations. That's a permissions test, not a brand test.

An enterprise-tier account with an admin who never turned off the default workspace sharing setting fails this just as badly as pasting a client's file into a free consumer tool. A firm using a paid product with a data processing agreement still has to check who inside that vendor, and who inside the firm, can actually reach the data. It lines up with what we found writing about whether AI prompts are discoverable: the protection questions that show up in litigation and the compliance question in this statute are asking the same thing from two different directions. Who can see it, and did you check.

Arbitrators and judges get their own rule, and it's stricter

SB 574 doesn't stop at lawyers. New Code of Civil Procedure section 1282.1 bars an arbitrator from delegating any part of the decision-making process to generative AI. The arbitrator has to take responsibility for the entire award and disclose, before relying on it, any AI-generated information that came from outside the evidentiary record. A separate new section, 180, expands the certification and complaint process the State Bar runs for alternative dispute resolution providers, giving it a hook to reach AI-related conduct. The September text also amended Business and Professions Code section 6173 to direct the Judicial Council to revisit its standards for judicial officers using generative AI. The signed-version summaries we found do not mention that change, so treat it as unconfirmed.

Put the two duties side by side and the arbitrator's is tighter than the lawyer's. A lawyer has to verify and correct. An arbitrator can't hand off the reasoning at all, full stop, and has to say so if AI material outside the record shaped the outcome.

What changes now that it is signed, and what does not

The governor signed, and most provisions take effect January 1, 2027, the standard start date for California legislation without an urgency clause. That leaves a few months to fix your settings and your filing habits before a judge can cite the new text.

The signed version also leaves two things open. As Holland & Knight notes, the statute does not define "delegate" or "the practice of law", so the line between AI assistance and AI delegation is unclear. And the court-disclosure duty does not say what form disclosure takes or how much AI involvement triggers it. Our read is to disclose whenever AI touched a filed document until a court or the State Bar says otherwise.

Here's the part worth sitting with. The statute changes less than the headlines suggest. It does not change what a California lawyer is already supposed to be doing under the Rules of Professional Conduct: keep client information confidential, stay competent with the tools you actually use, tell the court the truth, and supervise the staff who are already running client work through a chatbot. What the bill buys is a second, statute-level citation a judge can reach for during a sanctions motion instead of building the argument from a comment to Rule 1.1. That's a real difference in how fast a case against a lawyer moves. It is not a reason to wait for the effective date before you fix your firm's AI settings.

Our free guide walks through what to lock down before client material goes near a model, and the toolkit lists which tiers actually carry the access controls this article assumes you're checking.

FAQ