What the Law Requires

The core provisions appear in a new Section 6068.1 of the Business and Professions Code and in amendments to Code of Civil Procedure Section 128.7. Highlights include:

• No delegation. An attorney may not “delegate the practice of law” over to generative AI. AI may assist; it may not substitute for professional judgment.
• Confidentiality. An attorney may not enter confidential, personal identifying, or other nonpublic information into an AI system unless access to that information is limited to the attorney and people the attorney has authorized, who are themselves bound to protect it. The law lists examples, including dates of birth, medical information, financial information, and account numbers.
• Verification generally. An attorney must take reasonable steps to check the accuracy of AI outputs—including every case and statutory citation—and to correct any erroneous or “hallucinated” output in any material the attorney uses.
• Verification in court filings. A brief, pleading, motion, or other court paper may not contain any citation that “an attorney responsible for submitting the pleading” has not “personally verified.”
• Disclosure. An attorney must “disclose the use of generative artificial intelligence to the court for all documents submitted to the court,” and must consider disclosure when AI is used to create content for the public.
• Arbitrators. Arbitrators may not delegate any part of their decision-making to AI, and may not rely on AI-generated information outside the record without telling the parties first and, where practical, letting them comment.

Violations of the filing rule are enforced through California’s existing sanctions framework, which includes a 21-day window to withdraw or correct a challenged paper and makes law firms jointly responsible, absent exceptional circumstances, for violations by their partners, associates, and employees.

The Scope of the Law

The new law governs only “generative artificial intelligence,” which it defines as a system that “can generate derived synthetic content.” That includes common products such as Harvey, Legora, and Co-Pilot, but does not cover the use of non-generative artificial intelligence, such as recommendation or search engines, predictive analytics, or automated data processing tools.

It is also notable that while the law’s strictest obligations apply to filings in litigation, the law also requires that attorneys verify AI outputs more generally and that they “consider” when disclosure of the use of AI may be merited in other contexts. The law thus implicates legal practice beyond litigation—for example, in transactional work.

Two Questions the Text Leaves Open

First, what must be disclosed? Read literally, the disclosure clause (an attorney must “disclose the use of generative artificial intelligence to the court for all documents submitted to the court”) could require an attorney who has used AI for anything to say so in every document filed with any court. That is likely not what the legislature meant. The presumably better reading counsels disclosing when generative AI was used in preparing that particular document—perhaps only when AI contributed to the document’s content, research, or analysis (as opposed to more incidental uses like proofreading). Until courts or the State Bar weigh in, there is an unfortunate ambiguity.

Second, who must verify citations? The statute imposes obligations on “an attorney responsible for submitting the pleading.” But it leaves that role undefined. The presumably better reading of the non-specific phrase “an attorney responsible” likely permits any lawyer who works on the filing to verify citations. As ever, though, the attorney who signs the brief remains accountable as an ethical matter and should be prepared to demonstrate to the court that a compliant verification was performed. Notably, delegating the check to a second AI tool is unlikely to satisfy a requirement that “an attorney” verify cited authority.

What to do now

Clients should of course insist that their firms comply with these obligations (and may need to ensure their outside counsel guidelines do not discourage required checks). Moreover, within the company, clients should ensure that the tools they use internally are configured to keep confidential information appropriately protected, and that internal policy requires reasonable verification in appropriate circumstances.

We are monitoring the implementation of SB 574 and related developments in other states. For questions about how the law may affect your litigation, arbitration, or AI governance practices, please contact any member of our team.

This alert provides general information and is not legal advice.