Starting Jan. 1 , 2027, California lawyers must disclose AI use to the court and personally verify every citation they file

California's AI rules for lawyers go into effect on January 1, 2027.

Senate Bill 574, signed on September 30, does not ban the technology. It prohibits attorneys from delegating the practice of law to generative AI and requires those who use it in their practice to take reasonable steps to verify its output. Attorneys who use generative AI to assist in practicing law must also disclose that use to the court for documents they submit.

The bill also adds a citation rule to Code of Civil Procedure section 128.7, the statute on signing court papers, which carries its own sanctions. That rule follows a warning the Court of Appeal published in 2025, after it found that 21 of the 23 case quotations in a lawyer's opening brief were fabrications produced by AI tools. The court fined the lawyer $10,000.

Arbitrators get rules of their own, and the Judicial Council is told to revisit its AI standard for judges. This article sets out each rule from the chaptered text, sets the citation rule beside the court's sentence, and explains how the attorney duties are enforced. Passages that interpret the text, as distinct from reporting it, are marked as readings.

California's AI rules for lawyers take effect on January 1, 2027:

SB 574, by Senator Thomas Umberg (D-Santa Ana), is Chapter 858 of the Statutes of 2026. The bill history records a 39–0 Senate vote on January 29, a 75–0 Assembly vote on August 31 and a 40–0 Senate vote the same day to accept the Assembly's amendments. Governor Newsom signed it on September 30 with a group of other AI bills, about a year after he signed California's frontier AI safety law.

The bill contains no urgency clause and no delayed operative date, so it takes effect on January 1, 2027. It changes five sections of law:

• Business and Professions Code section 6068.1 (new): duties for attorneys who use generative AI.
• Code of Civil Procedure section 128.7 (amended): a citation rule for court filings, enforced through the section's sanctions procedure.
• Code of Civil Procedure section 1282.1 (new): limits on arbitrators.
• Code of Civil Procedure section 180 (new): a directive to the Judicial Council.
• Business and Professions Code section 6173 (amended): complaint procedures for State Bar-certified dispute resolution providers.

Business and Professions Code section 6068.1 sets four duties for attorneys:

The new section opens by saying that nothing in it abrogates an attorney's duty to exercise reasonable competence and diligence. Four duties follow. Duty 1 binds every attorney, and duties 2 to 4 apply to an attorney "who uses generative artificial intelligence to assist in the practice of law."

1. No delegation. "An attorney shall not delegate the practice of law to generative artificial intelligence."
2. Confidentiality. An attorney may not enter confidential, personal identifying or other nonpublic information into a generative AI system unless access to that information is restricted to the attorney and to people the attorney has authorized, who must be under obligations to protect its confidentiality.
3. Verification. An attorney must take reasonable steps to verify the accuracy of AI outputs, "including, but not limited to, the accuracy of all case and statutory citations," and to correct "any erroneous or hallucinated output" in any material the attorney uses.
4. Disclosure. An attorney must "Disclose the use of generative artificial intelligence to the court for all documents submitted to the court," and must "consider whether to disclose" its use when it creates content provided to the public.

"An attorney shall not delegate the practice of law to generative artificial intelligence."
Business and Professions Code section 6068.1(a)(2), effective January 1, 2027

The section defines generative AI as "an artificial intelligence system that can generate derived synthetic content," including text, images, video and audio. Its list of personal identifying information is broad. It names driver's license numbers, dates of birth, Social Security numbers, criminal record identification numbers, financial information and account numbers. It also names medical or psychiatric information, the addresses and phone numbers of parties, victims, witnesses and court personnel, and "Any other content sealed by court order or deemed confidential by court rule or statute."

The court disclosure duty covers every document submitted to the court:

The disclosure sentence is short, and the statute adds nothing to it. Section 6068.1 does not say what form a disclosure takes, where in a filing it belongs or how much AI assistance counts as "use." The section also does not define "delegate" or "the practice of law."

The duty goes further than the guidance lawyers and judges already had. The State Bar's Practical Guidance on generative AI, first published on November 16, 2023 and updated in a version the Board of Trustees approved on May 14, 2026, tells lawyers to check whether a court's rules or orders require disclosure of AI use and to comply. The judiciary's own Standard 10.80, which covers judges, asks a judicial officer to consider disclosure when generative AI is used to create content provided to the public.

Mandatory disclosure was not in the bill the Senate passed in January. That version, as leginfo's comparison with the final text shows, said nothing about disclosure to courts. Its only disclosure duty for attorneys was to consider disclosing AI use "if it is used to create content provided to the public." The court disclosure duty and the no-delegation sentence both appear in the Assembly's July 2 version and not in the June 22 version.

The citation rule in section 128.7 follows the Noland opinion with four changes:

Code of Civil Procedure section 128.7 is California's statute on signing court papers. By presenting a paper to a court, "an attorney or unrepresented party" certifies, among other things, that its legal contentions are warranted by existing law or by a nonfrivolous argument for changing it. SB 574 adds a new paragraph on citations.

The source of that paragraph is Noland v. Land of the Free, L.P., which the Second District Court of Appeal, Division Three, filed on September 12, 2025. The appeal concerned employment claims and was, in the court's words, "in most respects, unremarkable." The court wrote that it published the opinion because "nearly all of the legal quotations in plaintiff's opening brief, and many of the quotations in plaintiff's reply brief, are fabricated."

The opinion is specific about what went wrong, and the proportions matter. In the court's count, "appellant's opening brief contains 23 case quotations, 21 of which are fabrications." The court added that "most of the cases to which the quotes are attributed exist, the quotes do not," that many cited cases did not discuss the topics for which they were cited, and that "a few of the cases do not exist at all." The problem was mostly invented quotations attached to real cases.

The plaintiff's attorney told the court at oral argument that he wrote initial drafts, "enhanced" them with ChatGPT and ran the results through other AI tools to check for errors. He admitted that he did not read the enhanced briefs before filing them. The court imposed a $10,000 sanction payable to its clerk, ordered the attorney to give the opinion to his client and directed the clerk to forward it to the State Bar.

No California court had addressed AI-fabricated authority before, the court said, and so it published the opinion "as a warning." The Senate Judiciary Committee's analysis of SB 574 says the bill models its provisions on Standard 10.80 and on that ruling "regarding verifying cases and citations used in documents submitted to the courts." The analysis cites the ruling as 114 Cal.App.5th 426. This is the same method California Today applied to California's frontier AI law and the report from which it was drawn: set the two texts side by side.

The Noland warning and the new citation rule compare as follows:

The statute follows the opinion closely, and it differs in four places:

1. Advice became a mandate. The court wrote that no paper "should contain" unverified citations. The statute says a paper "shall not contain" them.
2. The AI clause moved and changed form. The court's sentence covered citations "whether provided by generative AI or any other source." The statute drops "or any other source" and covers "any citations," adding "including any citation provided by generative artificial intelligence."
3. "The attorney" became "an attorney."
4. "Read" was dropped. The court's sentence covered citations the responsible attorney "has not personally read and verified." The statute says "personally verified."

The last two changes were made late. The version the Senate passed in January said "the attorney" and "personally read and verified," which left two differences from the court's sentence. The Assembly's August 13 amendments produced the final wording, as leginfo's comparison of the July 2 version with the final text shows.

Two of the four changes carry legal weight, and what follows is a reading of the text, not a report of it. "Should" in an opinion is a court's instruction to the bar; "shall not" in a statute is a prohibition whose breach triggers the section's sanctions procedure. The removal of "read" leaves room to argue that an attorney can verify a citation without reading the cited case in full, for example by confirming that a quotation appears on the cited page. The statute does not define "verified," and section 6068.1 separately requires reasonable steps to verify "the accuracy of all case and statutory citations." How much reading verification requires is a question the courts will answer.

California’s "TFAIA" frontier AI law copies its own playbook almost word for word

California’s 2025 Frontier AI Policy Report laid the groundwork for SB 53, the Transparency in Frontier AI Act. Here’s how the state’s landmark AI safety law mirrors the Governor’s recommendations on transparency, risk, and whistleblower protections.

California TodayMac Douglass | Editor

Section 128.7 sanctions reach unverified citations from January 1:

Because the citation rule sits in subdivision (b) of section 128.7, the section's existing sanctions procedure applies to it. A court that finds a violation of subdivision (b) may sanction "the attorneys, law firms, or parties" responsible. The procedure has these features:

• A 21-day window. A party seeking sanctions must serve its motion before filing it. The motion may not be filed if the challenged paper is withdrawn or appropriately corrected within 21 days after service, or any other period the court sets.
• The court's own motion. A court may order an attorney, law firm or party to show cause. The order allows 21 days from its service to withdraw or appropriately correct the challenged paper.
• Firm responsibility. "Absent exceptional circumstances, a law firm shall be held jointly responsible for violations committed by its partners, associates, and employees."
• A deterrence limit. A sanction is "limited to what is sufficient to deter repetition" of the conduct.
• A discovery exclusion. The section does not apply to disclosures, discovery requests, responses, objections or discovery motions.

From January 1, one cross-reference points somewhere new. Subdivision (d)(1) says: "Monetary sanctions may not be awarded against a represented party for a violation of paragraph (2) of subdivision (b)." In the section as it stands today, paragraph (2) of subdivision (b) is the certification that legal contentions are warranted. SB 574 renumbers that certification as (b)(1)(B) and places the citation rule at (b)(2). The sentence in (d)(1) is unchanged, so it will then refer to the citation rule. The Legislative Counsel's digest does not mention the change.

As a reading: the result is sensible for citations, since a client does not check a lawyer's cases. Whether the Legislature also meant to remove the same protection for represented parties whose lawyers advance unwarranted legal contentions is not answered by the text.

A second open point concerns people who represent themselves. The existing certification binds "an attorney or unrepresented party." The new citation paragraph names "an attorney responsible for submitting the pleading" and says nothing about unrepresented parties. On that wording, the citation rule is addressed to attorneys. Self-represented litigants remain bound by the existing certifications, including the one on legal contentions.

Violations of section 6068.1 may lead to State Bar discipline:

Section 6068.1 has no penalty clause. The bill's text attaches no fine, no private right of action and no damages remedy to its four duties. The section instead sits in the State Bar Act, the body of law under which California attorneys are licensed and disciplined.

That placement matters because of Business and Professions Code section 6103, which provides that "any violation of the oath taken by him, or of his duties as such attorney, constitute causes for disbarment or suspension." Read plainly, and this is analysis, a duty that the Legislature writes into the State Bar Act is one of an attorney's duties for the purposes of that section. Whether a given lapse warrants discipline is for the State Bar Court and, ultimately, the California Supreme Court.

Court sanctions and Bar discipline are already linked. Under section 6086.7, a court must notify the State Bar of judicial sanctions against an attorney, other than sanctions for failure to make discovery and monetary sanctions under $1,000, and the State Bar must investigate. Under section 6068, an attorney must report such a sanction to the State Bar in writing within 30 days of learning of it. The Noland court used section 6086.7 when it ordered its clerk to forward the opinion.

The State Bar has been acting on AI misuse without the new statute. A staff report to its Board of Trustees dated May 14, 2026 records that the Bar issued an advisory on April 13 describing three discipline matters that involved AI misuse and fabricated authorities in court filings. The Bar has also proposed comment amendments to six Rules of Professional Conduct, including a comment to rule 3.3 on verifying cited authorities. After two public comment periods, the second of which closed on August 6, 2026, the Bar's Committee on Professional Responsibility and Conduct approved the proposals on September 11. A staff report for the Board's November 19 meeting proposes further changes, including a cross-reference to section 6068.1, and asks the Board to release the amended proposals for another 45-day public comment period. Rule amendments the Board adopts go to the California Supreme Court for approval.

Arbitrators cannot delegate any part of their decisionmaking to generative AI:

SB 574 adds section 1282.1 to the California Arbitration Act. It is a new section, and it leaves section 1282 as it was. An arbitrator, the new section says, "shall not delegate any part of their decisionmaking process to any generative artificial intelligence tool." An arbitrator also may not rely on AI-generated information from outside the record "without making appropriate disclosures to the parties beforehand and, as far as practical, allowing the parties to comment on its use."

Earlier versions asked more of arbitrators. The August 13 amendments removed four provisions, which said that:

• AI use must not replace an arbitrator's independent analysis of the facts, the law and the evidence;
• an arbitrator must not relinquish decisionmaking powers to AI, and must avoid delegating tasks to AI tools where that could influence procedural or substantive decisions;
• if an AI tool cannot cite independently verifiable sources, an arbitrator must not assume that such sources exist or are accurately described;
• an arbitrator takes responsibility for every aspect of an award.

The same amendments added a clause saying that nothing in the section abrogates an arbitrator's duty of neutrality or duty to exercise reasonable competence and diligence. The Assembly Privacy and Consumer Protection Committee's July analysis lists the American Arbitration Association as opposed unless the bill was amended.

The amendment to section 6173 is not about AI. It rewrites the complaint procedures for the State Bar's voluntary certification program for dispute resolution providers, makes those proceedings confidential and bars a voluntary association of attorneys from serving as the approved complaint body.

The Judicial Council must revisit Standard 10.80, its AI standard for judges:

The June 22 version of SB 574 would have required every judicial officer, in civil and criminal matters, to disclose whether anyone involved in researching or drafting a ruling "relied on generative artificial intelligence in that process." That requirement did not survive. The final section 180 says the Judicial Council "shall publicly revisit, and revise as necessary," Standard 10.80 "to reflect the further development of generative artificial intelligence."

Standard 10.80, adopted effective September 1, 2025, gives judicial officers guidance on using generative AI in their adjudicative work, including keeping nonpublic information out of public AI systems and verifying output. Its companion, California Rule of Court 10.430, required any court that allows generative AI to adopt a use policy by December 15, 2025. Section 180 sets no deadline for the Judicial Council's review.

The Assembly's August amendments removed two attorney duties from SB 574:

When it passed the Senate, the attorney section carried two duties that the final law does not. The August 13 amendments, made in the Assembly Appropriations Committee, removed both:

• Discrimination. A duty not to use generative AI to unlawfully discriminate against, or disparately impact, people on the basis of a long list of protected characteristics.
• Biased content. A duty to take reasonable steps to remove biased, offensive or harmful content from AI outputs.

The confidentiality duty had already changed in July. The June 22 version defined a "public generative artificial intelligence system," and the July 2 version replaced that defined term with the access test now in the duty. The August 13 amendments then changed the test's "public artificial intelligence system" to "generative artificial intelligence system." Both removed duties have counterparts in Standard 10.80, which the Senate Judiciary analysis names as a model for the bill.

California's AI rules for lawyers leave three points for courts and the State Bar to settle:

SB 574 fixes the principle and leaves the practice to others. Three points are unsettled as January 1 approaches, and identifying them is analysis:

1. The form of a court disclosure. The statute requires disclosure for all documents submitted to the court and prescribes no wording or placement. Courts can fill that gap by local rule, and the Judicial Council can do so statewide.
2. The line between assistance and delegation. The statute regulates attorneys who use AI "to assist in the practice of law" and forbids delegating the practice of law to it. It defines neither term.
3. The meaning of "personally verified." The Legislature dropped "read" from its own bill's earlier wording, which had followed the court's, and from January 1 section 128.7 carries the shorter phrase.

The Judicial Council's review of Standard 10.80 and the State Bar's pending rule comments are the next places to look for answers. California's AI rules for lawyers are in force from January 1, 2027.