On 30 September 2026 California Governor Gavin Newsom signed Senate Bill 574, making California the first state in the country to regulate lawyers' use of generative artificial intelligence by statute, rather than leaving the question to bar guidance or to case-by-case court sanctions. The law adds a new Section 6068.1 to the Business and Professions Code, amends Section 6173 of that code, and adds Sections 180 and 1282.1 to, and amends Section 128.7 of, the Code of Civil Procedure. It takes effect on 1 January 2027.
What SB 574 requires
- No delegating the practice of law. A lawyer may use generative AI to help with research, drafting, document review, discovery and correspondence, but the legal judgment behind the work has to stay with the lawyer.
- Confidentiality first. Confidential, personal or non-public information cannot go into an open, consumer-facing generative AI tool; tools restricted to the lawyer and authorized staff are allowed.
- Personal verification of every citation. Before filing a brief, the lawyer must take reasonable steps to confirm that each citation, whether AI-generated or not, actually exists, is still good law and supports what it is cited for.
- Disclosure in court filings. A filing prepared with the help of generative AI has to say so, and it still needs the signature of an attorney in their own name unless the party is self-represented.
- The same logic for arbitrators and judicial officers. They cannot hand off any part of their decision-making to generative AI, and must disclose any AI-generated information from outside the record that they relied on.
- A standing duty for the Judicial Council. It must keep the relevant judicial-administration standard under public review as generative AI keeps developing.
Why a statute, and why now
SB 574 arrives after a year in which courts on more than one continent were already sanctioning lawyers over AI-hallucinated citations, without needing a dedicated law to do it. In August, Galicia's Tribunal Superior de Xustiza fined a lawyer 1,800 euros for a brief built on invented case law, and referred the ruling to the Ilustre Colegio de Abogados de A Coruña for possible disciplinary action. In February, Colombia's Corte Suprema de Justicia, in Auto AC739-2026 of its Sala de Casación Civil, Agraria y Rural, sanctioned a lawyer for procedural recklessness after a petition relied on legal provisions and case law that did not match the official record. Both courts were careful to say that using AI is not the problem; filing unverified AI output is. California's law takes that same distinction and writes it into a lawyer's statutory duties, instead of leaving it to be worked out one sanction at a time.
What it means for a firm, starting now
Firms with any California-admitted lawyers have a little over two months to get three things in writing before 1 January: a citation-verification step that a specific person owns on every filing, a list of which AI tools are cleared for client-confidential input and which are not, and standard disclosure language for filings that used generative AI. None of that requires new technology. It requires deciding, on paper, who checks what, before the duty becomes a statutory one instead of a best practice.
At Lexpower, that division of labor is not a new compliance item to bolt on. It is the model itself: a specialist AI trained by practice area and by jurisdiction prepares the work against the official rule in force, and a licensed professional reviews, verifies and signs it before it reaches anyone. See how our specialists work.
Sources
- California Enacts Rules Governing Lawyers' Use of Generative AI — Holland & Knight
- California bill explicitly regulates attorneys' generative AI use — Canadian Lawyer
- Bill Text — SB-574 Attorneys, arbitrators, judicial officers, and alternative resolution providers — California Legislative Information