With Governor Newsom’s signing of SB 574, California has taken the first step in directly addressing how lawyers and arbitrators use generative artificial intelligence (“AI”). The bill adds Business and Professions Code section 6068.1, which prohibits attorneys from delegating the practice of law to generative AI and requires them to keep confidential, personally identifying, and other nonpublic information out of public generative AI systems. It also amends Code of Civil Procedure section 128.7 to require that attorneys personally read and verify the authorities cited in papers they sign, and it adds section 1282.1, which bars arbitrators from delegating decision-making to generative AI. While much of this codifies the duties of competence, confidentiality, and candor that lawyers already owe, it also provides a statutory hook for discipline and sanctions related to the misuse of AI.
Unfortunately, SB 574 also includes a provision that creates a problem, rather than solves it.
Business and Professions Code Section 6068.1(a)(3)(C) will now require an attorney who uses generative AI to assist in the practice of law to “disclose the use of generative artificial intelligence to the court for all documents submitted to the court.” The statute does not say how the disclosure must be made, where it must appear, or what constitutes “use” that triggers disclosure.
The definition of AI makes matters worse. The statute defines generative AI as “an artificial intelligence system that can generate derived synthetic content, including text… that emulates the structure and characteristics of the system’s training data.” Put simply, AI is any software that: (1) uses a trained model, (2) produces output the user did not directly author, and (3) does so by learning patterns from data. While the bill’s authors may have been thinking about ChatGPT-type chatbots, the broad definition sweeps in far more.
The litigator who uses the AI-assisted search and summarization features in Westlaw and Lexis receives output based on patterns learned by their respective models. But so do everyday writing tools such as Microsoft Editor and Grammarly. Since every Google search result starts with an AI-generated response, does looking anything up on Google while drafting a brief trigger a disclosure for “use of generative artificial intelligence”? Does the use of technology-assisted review systems in discovery, whether machine learning that searches through millions of documents for responsive materials or even assistive search systems like CS Disco’s Cleo, require disclosure when discussing a document produced in e-discovery? Nothing in the text limits the disclosure duty to tools that replace an attorney’s exercise of professional judgment (e.g., whether a case supports the proposition for which it is being advanced) with AI.
The likely result is reflexive, boilerplate disclosure. Cautious practitioners will list their entire technology stack in every filing: “Under the broad definition of generative AI in Business and Professions Code section 6068.1, the undersigned used the following software in the practice of law: Microsoft Word, Westlaw, Grammarly, Microsoft Copilot, Microsoft Outlook, Disco,” and so on. A disclosure that every filer makes about nearly every tool tells the court nothing.
That raises a more basic question: what is the disclosure for? If the goal is to alert the judge that the cited cases may not support the stated propositions, the statute is redundant. Attorneys must already ensure that their authorities exist and that the brief accurately reflects their holdings, and they face sanctions under Code of Civil Procedure section 128.7 and the court’s inherent authority when they do not. Judges, for their part, are expected to read the cases themselves because, well before the advent of computers, lawyers miscited cases. A lawyer who submits hallucinated citations is sanctionable whether or not a disclosure appears on the signature page, and a lawyer who verifies every citation has not become less reliable because Grammarly encouraged them not to end their sentence with a preposition.
The Legislature was concerned about a real problem, but the existing rules already address it. What the disclosure mandate adds is ambiguity, compliance risk for lawyers who guess wrong about what counts, and, in all likelihood, a filing ritual with no informational value. Ironically, the solution to the problem the Legislature has created is the very mechanism that should have been left to address the issue in the first place. Local Rules will need to be updated (ideally before SB 574 becomes effective on January 1, 2027) to focus disclosures on what may actually be important to Judges. Of course, the only disclosure that should matter is that the filer is a member of the bar, as our ethical duties already govern our use of technology.
The legal profession’s approach to AI is still being written, and California’s SB 574 may be just the beginning. Greenspoon Marder’s Technology, Privacy & Data Security practice group closely monitors legal and regulatory developments affecting artificial intelligence, privacy, cybersecurity, and digital innovation. Visit our practice group page to explore our insights and learn how we help clients navigate an increasingly complex technology landscape.
This publication is provided by Greenspoon Marder LLP is issued for informational purposes only and is not intended to be construed or used as general legal advice nor a solicitation of any type. Please contact the author(s) or your Greenspoon Marder LLP contact if you have any questions regarding the currency of this information. The hiring of a lawyer is an important decision. Before you decide, ask for written information about the lawyer’s legal qualifications and experience.
