California has put lawyers’ use of generative AI into statute.
Governor Gavin Newsom signed SB 574 on September 30, 2026, introducing rules for attorneys and arbitrators. Reuters describes it as the first U.S. state law specifically targeting these professionals’ use of the technology. The rules take effect January 1, 2027, under California’s ordinary legislative timetable.
For Canadian lawyers, the development offers a concrete comparison: how far should AI oversight depend on professional guidance, and when should legislation prescribe the steps?
What SB 574 requires
The enacted law establishes several obligations:
- No delegation of legal practice to generative AI.
- Reasonable verification of AI output, including case and statutory citations, and correction of erroneous or hallucinated material.
- Restricted access to confidential inputs: protected information cannot be entered into systems unless access is limited to the attorney and authorized people bound to protect confidentiality.
- Disclosure to the court of generative AI use for documents submitted; lawyers must also consider disclosure for AI-created public content.
- Personal verification of every citation in a court filing by the responsible attorney - not just AI-generated citations.
Arbitrators cannot delegate any part of decision-making to generative AI. Before relying on AI-generated information outside the record, they must make appropriate disclosures and, as far as practical, allow party comment.
Why personal citation verification matters
Analysis: For a firm, the operational question is who performs the final review.
A policy that says “check AI output” leaves several matters unresolved. Does checking mean confirming that a judgment exists? Reading the cited passage? Confirming that the proposition survives later authority? Who records completion before submission?
A useful verification process should distinguish those tasks. Finding a real case with the correct name is only the beginning of evaluating whether it supports an argument.
Firms reviewing their procedures could assign a responsible reviewer, preserve source links and record unresolved questions. These are suggested workflow measures, rather than additional requirements specified by SB 574.
Canadian lawyers already have an AI responsibility framework
The Law Society of Ontario’s April 2024 white paper addresses generative AI through existing professional obligations. It discusses competence, confidentiality, supervision and verification, and recommends firm policies and product due diligence. It also directs licensees to check the requirements of the particular court receiving their documents.
California’s statute does not establish Canadian requirements. Its value here is comparative: it gives regulators and practitioners specific wording to examine.
What law firms should review before 2027
Analysis: Start with the journey of one document through the firm.
Identify where AI enters the process, what information it receives, who reviews the output and how the firm determines whether disclosure is necessary. Include work prepared by junior lawyers, staff and outside service providers.
This exercise can reveal gaps that a list of approved products misses. A tool may be approved for one task while a particular use still requires closer review.
For firms working with California counsel, discuss responsibility early enough to accommodate the applicable filing requirements. Avoid discovering at submission that the document’s preparation history is unclear.
The practical question is straightforward: Can the firm explain who checked the work, what they checked and which rules governed its use?
Primary source: Read the enacted SB 574
Related coverage: Sheppard Mullin Data Breach Lawsuit Highlights Social-Engineering Risk.