California SB 574: The First State Law Telling Lawyers How to Use AI

California has become the first state in the nation to regulate attorneys' use of generative artificial intelligence by statute. Senate Bill 574, signed by Governor Gavin Newsom on September 30, 2026, takes effect on January 1, 2027. It does not prohibit lawyers from using generative AI. It sets the standards they must meet when they do — and it adds a personal citation-verification rule that applies whether or not AI was involved.

The law is significant because it is first. Courts and bar associations have been issuing guidance and sanctions for AI misuse for three years, ever since lawyers began filing briefs containing cases invented by chatbots. SB 574 moves those expectations from guidance into statute.

What the Law Requires

SB 574 has two parts. The first adds a new section to the State Bar Act, Business and Professions Code Section 6068.1, setting out duties for any attorney who uses generative AI to assist in the practice of law. The second amends the sanctions statute, Code of Civil Procedure Section 128.7, to require personal verification of citations.

The duties in Section 6068.1 are:

No delegation of the practice of law. A lawyer "shall not delegate the practice of law to generative artificial intelligence." The statute permits AI to "assist in the practice of law" but prohibits delegating it — and does not define where permissible assistance becomes prohibited delegation. That line will be drawn case by case, and it is the provision most likely to generate disputes.

Confidentiality. A lawyer may not enter confidential, personally identifying or other nonpublic information into a generative AI system unless access to what is input "is restricted to the attorney and persons authorized by the attorney under obligations to protect the confidentiality of the information." In practice, that means the free, consumer-grade chatbot is off limits for client material; only tools with access controls and confidentiality protections qualify.

Verification and correction. A lawyer using generative AI must take reasonable steps to verify its output and correct erroneous or hallucinated material they use. This is the statutory version of the lesson from the hallucination cases: the lawyer, not the model, owns the work product.

Disclosure. A lawyer must disclose the use of generative AI to the court for all documents submitted to the court, and must consider whether to disclose it when generative AI is used to create content provided to the public. The statute does not specify the form of disclosure or the minimum level of AI involvement that triggers it — both are open questions that courts and the State Bar will have to resolve.

The citation rule is the sharpest part of the law. Under the amended Section 128.7, no court filing may contain a citation the responsible attorney has not personally verified, including any citation provided by generative AI — and the rule applies whether or not AI was used. Violations can bring sanctions, including attorney's fees and penalties, and law firms are jointly responsible absent exceptional circumstances. The Section 6068.1 duties are also enforceable through State Bar discipline.

What Counts as Generative AI?

The statute defines "generative artificial intelligence" as an AI system that "can generate derived synthetic content… that emulates the structure and characteristics of the system's training data."

The law also reaches beyond attorneys in court. It addresses arbitrators' use of generative AI, judicial guidance, and the State Bar's voluntary ADR-provider certification programme.

The Wider Picture: New York and the United Kingdom

California is first, but it is not alone. New York has a pending proposal, Senate Bill S2698 (sponsored by Senator Hoylman-Sigal, with Assembly companion A8546), that would require a separate affidavit disclosing generative AI assistance in drafting a court filing and certifying that a human reviewed the source material and verified the accuracy of the generated content. The bill, introduced in January 2025, is currently in the Senate Rules Committee after being committed there on June 13, 2025. Unlike California's law, it does not expressly require that the attorney responsible for the filing perform the verification, and it contains no enforcement mechanism for non-compliance beyond perjury liability for a knowingly false certification.

Across the Atlantic, the Civil Justice Council of England and Wales issued its Final Report on the Use of AI for Preparing Court Documents on October 29, 2026, with 15 recommendations and a more restrained, risk-based approach: no new AI declaration for advocacy documents, but a proposed declaration that AI has not generated or influenced the substantive content of witness statements of fact, and disclosure by experts where AI generated or influenced their reports. The recommendations are not yet rules; the Civil Procedure Rule Committee must take them forward, with no timetable set.

Why It Matters for Founders and Investors

SB 574 is a compliance deadline and a market signal at the same time. Every California law firm — and every company that sells AI tools to lawyers — has until January 1, 2027 to build the workflows the statute requires: access-controlled AI tools, verification checklists, citation-checking processes and disclosure templates. That is a product opportunity for legal-tech founders, and a diligence question for investors: does the tool keep client data inside a restricted environment, does it log what was verified and by whom, and does it generate the disclosure the court will expect?

Angeled.com connects founders with investors.