A lawyer may use generative AI. A lawyer may not let it practice law, may not feed it a client’s secrets through an open door, and may not file its citations unread.
On September 30, 2026, Governor Newsom signed Senate Bill 574 (Stats. 2026, ch. 858), widely reported as the first state statute in the country to regulate how attorneys and arbitrators use generative artificial intelligence. It takes effect January 1, 2027. Most of what it requires was already the profession’s understanding of its duties of competence, confidentiality and candor. What changes is that those duties are now written into the State Bar Act and the Code of Civil Procedure, in terms specific enough to enforce.
For clients, the statute is a useful measure of what to expect from counsel. This post covers what it requires and the questions it is reasonable to ask.
Why the Legislature acted
Generative AI tools write fluent, confident legal prose, and they sometimes invent the authority that prose relies on. In Noland v. Land of the Free, L.P. (2025) 114 Cal.App.5th 426, the Court of Appeal found that 21 of the 23 case quotations in an opening brief were fabricated, sanctioned the attorney $10,000, and referred the matter to the State Bar. It was the first published California opinion on the problem.
The courts had already set rules for themselves. Since September 1, 2025, rule 10.430 of the California Rules of Court and standard 10.80 of the Standards of Judicial Administration have governed the use of generative AI by courts and judicial officers. SB 574 extends that structure to the lawyers and arbitrators who appear in front of them.
What the statute requires of attorneys
SB 574 adds section 6068.1 to the Business and Professions Code. It contains one prohibition and three duties.
No delegation. “An attorney shall not delegate the practice of law to generative artificial intelligence.” The statute does not define the line between assistance and delegation, but its structure answers most of the question: the tool may assist, and the attorney’s own duty of “reasonable competence and diligence” is expressly preserved. The judgment has to remain the lawyer’s.
Confidentiality. An attorney may not enter “confidential, personal identifying, and other nonpublic information” into a generative AI system unless access to what the attorney inputs is restricted to the attorney and to persons the attorney has authorized who are themselves under obligations to protect it. The practical distinction is between a consumer chatbot, whose provider may retain and use what is typed into it, and a tool supplied under contract terms that keep the firm’s inputs confidential. The definition of personal identifying information is broad. It includes dates of birth, Social Security and driver’s license numbers, medical and financial information, account numbers, anything sealed or made confidential by court order, rule or statute, and the addresses and telephone numbers of parties, victims, witnesses and court personnel.
Verification. An attorney must take reasonable steps to verify the accuracy of AI output, “including, but not limited to, the accuracy of all case and statutory citations,” and to correct “any erroneous or hallucinated output in any material used by the attorney.” This duty is not limited to court filings. It reaches advice letters, contracts, and anything else the attorney uses.
Disclosure. An attorney must disclose the use of generative AI “to the court for all documents submitted to the court.” The statute does not prescribe a form or say how much detail is required, and courts can be expected to fill that in by local rule or standing order. For content provided to the public, the attorney must “consider whether to disclose” it.
A second rule, with sanctions attached
Separately, SB 574 amends Code of Civil Procedure section 128.7, the statute under which an attorney’s signature certifies that a filing has legal and factual support. New subdivision (b)(2) provides that a brief, pleading, motion or other paper filed in any court “shall not contain any citations that an attorney responsible for submitting the pleading has not personally verified, including any citation provided by generative artificial intelligence.”
Two features are worth noticing. The rule is not limited to AI: a citation copied from a treatise, a prior brief or a colleague’s memo has to be personally verified too. And it carries section 128.7’s existing remedies, which include nonmonetary directives, a penalty paid to the court, and the opposing party’s attorney’s fees, with a law firm ordinarily held jointly responsible for its lawyers’ violations. As amended, the section also bars monetary sanctions against a represented party for a violation of subdivision (b)(2). The consequence of an unverified citation falls on the lawyer and the firm, not on the client.
Arbitrators and the courts
New Code of Civil Procedure section 1282.1 provides that an arbitrator “shall not delegate any part of their decisionmaking process to any generative artificial intelligence tool,” and may not rely on AI-generated information from outside the record without first disclosing it to the parties and, as far as practical, allowing them to comment. For any business or agency whose contracts send disputes to arbitration, that is a new and specific assurance about who is deciding the case.
The bill also directs the Judicial Council to revisit standard 10.80 as the technology develops, and revises the State Bar’s complaint procedures for certified alternative dispute resolution providers.
What this means for public agencies
A public agency’s lawyers hold information that is nonpublic by law: closed-session discussions, personnel records, pending litigation strategy, real property negotiations. Beginning in January, a city attorney or outside counsel who puts that material into an AI tool without the access restrictions the statute describes violates the State Bar Act. Agencies that are adopting their own AI policies for staff should make sure the policy and the legal department’s practice are consistent, and should expect their outside counsel to be able to describe the tools they use.
Questions clients should ask
Does the firm use generative AI on my matters, and for what tasks?
Which tools, and under what terms is the information entered into them kept confidential?
Who checks the output, and does an attorney read every authority cited before a document is filed or sent?
How will the firm handle the disclosure to the court that the statute requires?
If my contract has an arbitration clause, what does the provider’s policy say about arbitrators’ use of AI?
None of these questions should be difficult for a lawyer to answer.
Looking ahead
SB 574 leaves real questions open: what counts as delegation, what form disclosure to the court must take, and how the State Bar will enforce a section that has no penalty provision of its own. Those will be answered over the next few years by local rules, discipline decisions and sanctions orders. The direction is already clear. The tool is permitted, and the responsibility for everything it produces stays with the attorney who signs.
