California’s SB 574 Asks Who Can Read What a Lawyer Puts Into AI
Governor Newsom signed Senate Bill 574 (Umberg) on September 30, 2026. It is Chapter 858, Statutes of 2026, and takes effect January 1, 2027. The law regulates how California lawyers use generative artificial intelligence. The Legislature passed it on August 31, 2026, and the provisions quoted below are the text as signed.
Under the law, a lawyer may put nonpublic information into an AI tool only if everyone who can access it was authorized by the lawyer and is bound to keep it confidential.
The new California statute codifies confidentiality requirements for every AI system
SB 574 adds section 6068.1 to the Business and Professions Code. Subdivision (a)(3)(A) provides that an attorney shall:
“Not enter confidential, personal identifying, and other nonpublic information into a generative artificial intelligence system for which access to confidential, personal identifying, or other nonpublic information the attorney inputs into the system is not restricted to the attorney and persons authorized by the attorney under obligations to protect the confidentiality of the information.”
The provision covers more than client confidences. The text does not limit “other nonpublic information” to information about a client. Read literally, it reaches a witness’s home address or an opposing party’s financial records produced in discovery.
The August 13 amendments struck the word “public” from the provision, so the test applies to every generative AI system. A paid product built for lawyers is subject to it the same as a free chatbot.
Five questions to ask about any AI tool
California’s new statutory test can be put to any tool, whether ChatGPT, Claude, Gemini, or a product built for lawyers, as five questions.
- Which companies receive my client’s information? Many legal AI products run on another company’s model, and a third company may host it.
- Does each company keep prompts and responses, and for how long?
- Can a person at any of those companies read them, and in what circumstances?
- What confidentiality obligation binds that person?
- Where is each answer written in a contract that applies to my account?
While consumer-facing AI tools from OpenAI, Anthropic, Google, and others might allow customers to opt out of training, successfully opting out of training does not resolve all confidentiality issues. Additionally, opting out of training on these tools can only apply to one account at a time. A law firm managing partner cannot easily confirm that every lawyer at the firm has correctly configured ChatGPT or Claude, nor can they prevent the provider from changing the setting’s default or location in a future update.
Large law firms address this challenge by contract. A commercial contract puts no-training, retention, and access terms in one document that covers every user at the firm, and a lawyer there can answer the five questions by reading it.
How Bells Up AI answers the five questions
Bells Up AI has built its product for the solo and small-firm lawyers who want to use AI while satisfying these obligations. Every model on the Bells Up platform is reached through commercial contracts that bar training on your prompts and responses. Those terms apply to every plan, including our free plan, and to every user at a firm.
Our privacy reference page answers each of the five questions in detail and goes provider by provider, with quotations, links, and the caveats that apply to each.
Section 6068.1 also requires that the attorney authorize each person with access. Bells Up AI’s Terms of Service name every provider we use, so a lawyer sees the full list of companies before agreeing to anything. The providers’ obligations run to Bells Up AI under our agreements with them, and our obligation runs to the lawyer under our Terms of Service.
For more about how Bells Up AI protects client confidences, see our privacy reference page.