California’s SB 574: The First California AI Rules for Lawyers

California is close to putting generative AI obligations for lawyers into a statute. Senate Bill 574 (SB 574), authored by Senator Thomas Umberg, would add explicit duties for attorneys who use generative AI in the practice of law and would restrict certain uses of AI by arbitrators.[1]

SB 574 has already passed the California Senate unanimously (39–0) and is now in the Assembly, where it has been read for the first time and is currently held at desk.

The bill does not try to stop lawyers from using AI. Instead, it makes the lawyer (and, in arbitration, the arbitrator) legally responsible for managing predictable risks such as confidentiality leakage, fabricated citations, and biased or discriminatory outputs.

Why the Legislature Thinks This Is Necessary

Generative AI is already being used for legal research, drafting, summarizing records, and client communications. The efficiency gains are real, but so are the risks. AI can produce confident but false answers, invent citations, and recycle or expose sensitive information depending on how it is used.

California courts have already dealt with AI-driven filing errors. In a published decision, Noland v. Land of the Free, L.P., the Court of Appeal sanctioned counsel after a brief included fabricated quotations and other AI-generated errors, emphasizing that lawyers must personally read and verify what they cite. [2]

Until now, the core guardrails have come from professional responsibility principles and State Bar guidance.[3] SB 574 would move key expectations from “best practice” into “black letter” law.

SB 574 would add Business & Professions Code section 6068.1 and impose four practical obligations on any attorney using generative AI in the practice of law:

1. Keep Confidential and Nonpublic Data Out of Public AI Systems

An attorney would have to ensure that confidential, personally identifying, or other nonpublic information is not entered into a public generative AI system. In plain English: do not paste client facts, documents, or identifiers into consumer tools unless you have clear protections on retention, training, access controls, and confidentiality.  Most of the paid subscriptions to ChatGPT, for example, provide a “private” space and promises the input, and output, will not be used outside of that closed environment or to train its AI models. The free versions of ChatGPT, however, do not offer this kind of privacy and client protection environment. With AI, you get what you pay for.

2. Avoid Unlawful Discrimination and Disparate Impact

An attorney would have to ensure the use of AI does not unlawfully discriminate or disparately impact individuals or communities based on protected characteristics. If AI is used to sort, screen, prioritize, or recommend actions, bias risk becomes an express compliance issue.

3. Take Reasonable Steps to Verify Accuracy and Remove Harmful Content

An attorney would have to take reasonable steps to verify accuracy, correct hallucinated or erroneous output, and remove biased, offensive, or harmful content in AI material used. The output is not the end of the job; it is the start of the review.

4. Consider Disclosure When AI-Generated Content Is Provided to the Public

The bill would not mandate disclosure across the board, but it would require lawyers to consider whether disclosure is appropriate when AI is used to create content provided to the public (for example, marketing, public-facing advisories, or other broad communications).  Our firm, as an example, updated our standard client retention agreements to specify how and when we use AI and we assure our clients it not only makes our attorneys and staff more productive, but we do not charge for the actual use of AI. We also clearly disclose our AI policy on our website.

The Sharpest Edge: Filings and Citations

SB 574 would also amend Code of Civil Procedure section 128.7 (California’s sanctions statute for improper filings). It would add an explicit requirement: no brief, pleading, motion, or other paper filed in court may contain citations the attorney has not personally read and verified, including citations generated by AI.

This is the provision that will change behavior fastest. It turns what many judges already expect into a statutory bright line: if it is cited to the court, the lawyer is on the hook for it.

AI and Arbitration: No Outsourcing the Decision

SB 574 would add Code of Civil Procedure section 1282.1, regulating arbitrators’ use of generative AI. The theme is straightforward: AI may not become a silent decision-maker.

• An arbitrator could not delegate any part of the decision-making process to generative AI, and AI could not replace the arbitrator’s independent analysis of facts, law, and evidence.
• If an arbitrator intends to use AI-generated information outside the record, the arbitrator would have to disclose it in advance and, as far as practical, give the parties an opportunity to comment.
• If an AI tool cannot cite independently verifiable sources, the arbitrator may not assume those sources exist or that the AI’s characterization is accurate; the arbitrator remains responsible for the award.

If enacted, SB 574 will push firms toward clearer policies and better documentation of AI workflows. Three near-term implications are worth highlighting:

1. Tool Selection Becomes a Professional Responsibility Issue

Lawyers will need to distinguish between public consumer tools and systems with enterprise-grade confidentiality and retention controls. “We didn’t know how the tool handled data” will not be a good answer.  Attorneys must know and must ensure client data is not disseminated outside of a secure work area.

2. Verification Becomes a Defined Step in the Work Product Pipeline

Firms should expect to implement checklists and supervision rules, citation pulls, record checks, and required human review before anything goes to a client or court.

3. Litigation Risk Increases When AI Is Used Casually

A single hallucinated authority can now lead more directly to sanctions exposure, reputational damage, and malpractice claims. AI can speed up drafting, but it can also accelerate mistakes.

What Clients Should Ask and Why It Matters

Clients should not need to fear AI, but they should care how it is used. A few direct questions can prevent misunderstandings and reduce risk:

• Are you using generative AI on my matter, and if so, for what tasks?
• What protections prevent my confidential information from being entered into a public AI system?
• What is the human review process before AI-assisted work product is delivered or filed?
• How do you handle billing when AI increases speed, especially in hourly matters?

For some matters, clients may also want to address AI use explicitly in engagement letters or outside counsel guidelines, especially around confidentiality, acceptable tools, and retention of matter data.

As of March 24, 2026, SB 574 has cleared the Senate unanimously and has moved to the Assembly. It has been read for the first time in the Assembly and is currently held at desk. Next steps are committee referrals and hearings, followed by an Assembly floor vote. If amended, the bill would typically return to the Senate for concurrence before going to the Governor.

SB 574 would make one point unavoidable: lawyers can use AI, but they cannot outsource judgment. If it passes, California will have a clear statutory framework that forces responsible AI use in the practice of law and gives courts and clients a cleaner yardstick to measure it.

Shustak Reynolds & Partners, P.C. focuses its practice on securities and financial services law and complex business disputes.
We represent many investment advisors, financial professionals, broker-dealers, registered representatives, investors and businesses.
Attorney Erwin J. Shustak can be reached in the firm’s San Diego office at (619) 696-9500.

[1] California Legislature SB 574 (text, history, and votes), https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260SB574

[2] Noland v. Land of the Free, L.P. (published Cal. Ct. App. opinion, Sept. 12, 2025),  https://courts.ca.gov/opinion/published-extended-post/2025-09-12/b331918

[3] State Bar of California Generative AI Practical Guidance, https://www.calbar.ca.gov/Portals/0/documents/ethics/Generative-AI-Practical-Guidance.pdf