Get started

AI and ethics

California's SB 574 puts lawyer AI rules into statute: what changes on January 1, 2027

Governor Newsom signed SB 574 on September 30, 2026. Starting January 1, 2027, California lawyers must disclose generative AI use in court filings, personally verify citations, and keep confidential information out of unsecured AI tools. Here is what it requires and how to prepare.

Published 5 min read

By Robert Pasha

On September 30, 2026, Governor Gavin Newsom signed SB 574, which puts rules for lawyers' use of generative AI into California statute. It takes effect January 1, 2027, according to Holland & Knight's October 1 alert. If you practice in California, you have about ten weeks to get your drafting, filing, and confidentiality habits in line.

Until now, most AI guidance for lawyers has come from bar opinions that explain how existing rules apply. SB 574 is different. It adds a new section to the Business and Professions Code and amends the statute that governs sanctions for court filings. The California courts newsroom summed up the headline change plainly: starting next year, California attorneys must disclose when they use generative AI to create documents submitted to a court.

What SB 574 requires

The law was chaptered as Stats. 2026, ch. 858, according to Morgan Lewis's October 2 LawFlash. Based on that summary and the Holland & Knight alert, the core obligations are these.

New Business and Professions Code section 6068.1

  • Assist, do not delegate. You may use generative AI to assist in the practice of law, but you may not delegate the practice of law to it. The statute does not define where assistance ends and delegation begins.
  • Protect confidential information. Holland & Knight reads the law to bar entering confidential, nonpublic, or personally identifying information into a generative AI system unless access is limited to the attorney and authorized people who are bound by confidentiality.
  • Verify the output. You must take reasonable steps to verify AI output, including case and statutory citations, and correct erroneous or hallucinated material.
  • Disclose to the court. You must disclose generative AI use for documents submitted to a court. Morgan Lewis also notes a duty to consider whether disclosure is appropriate for content provided to the public.

Amended Code of Civil Procedure section 128.7

The attorney responsible for a California court filing must personally verify its citations. This applies whether or not AI was used. Holland & Knight notes that research can still be delegated, but the responsible attorney cannot rely solely on someone else's check. Violations fall under the existing section 128.7 sanctions framework, and firms can be jointly responsible for their partners, associates, and employees.

Arbitrators and the Judicial Council

A new Code of Civil Procedure section 1282.1 bars arbitrators from delegating any part of their decision-making to generative AI. It also bars them from relying on AI-generated information outside the record unless they first disclose it to the parties. A new section 180 directs the Judicial Council to revisit its guidance on judicial AI use. Both points come from the Morgan Lewis summary.

The questions the statute leaves open

Both firm alerts flag the same gaps. Expect them to shape how the law works in practice.

  • What counts as "use." The statute does not say whether research, summarizing, proofreading, or editing with an AI tool triggers court disclosure. Morgan Lewis lists this as an open question.
  • What the disclosure looks like. There is no required form or location. Holland & Knight suggests document-specific disclosure as the more prudent course.
  • Where delegation starts. There is no test for how much attorney involvement a workflow needs. Morgan Lewis also raises how the law applies to more autonomous, "agentic" tools.

Until courts or the Judicial Council fill these gaps, the safe reading is the broad one. If a generative AI tool shaped the text of something you file, plan to disclose it.

Why this matters outside California

California is the first state to do this by statute, but it is not acting alone. On the federal side, the Tenth Circuit has proposed revisions to its Rule 46.5 that are, in the court's words, "intended to address the increased use of generative artificial intelligence by both lawyers and pro se litigants." Comments closed October 18, 2026, and the changes are set to take effect January 1, 2027, according to the court's public rules memo.

Meanwhile, guidance across the country remains uneven. A New York City Bar Association paper published in August reviewed AI guidance in 21 jurisdictions and found no single approach, Newsweek reported. If you practice in more than one state, build your process around the strictest rule you face. For many lawyers, that will soon be California.

What to do before January 1, 2027

  1. Inventory your AI tools. List every tool anyone in your office uses that generates text: chat assistants, drafting add-ins, research tools, transcription and summary features built into software you already pay for. You cannot comply with a disclosure rule if you do not know where AI touches your work.
  2. Check each tool's confidentiality terms. For each one, confirm who can access what you enter, whether the vendor keeps it, and whether it is used to train models. If you cannot confirm access is limited to you and people bound by confidentiality, keep client information out of it.
  3. Write down your citation check. The signing attorney should personally pull and read every cited authority before filing, whether a person or a tool found it. Keep a short note in the file showing it was done.
  4. Draft a standard court disclosure. Prepare one or two plain sentences you can adapt to each filing, and decide when you will use them. Watch for local rules or Judicial Council guidance that sets a form.
  5. Update your engagement letter. SB 574 focuses on courts, but clients should not first learn about your AI use from a disclosure on their own filing. A short paragraph on how you use AI and how you protect their information heads that off.
  6. Train your staff. Paralegals and assistants who prepare drafts need to know the rules too, since firms can share responsibility for sanctions.
  7. Ask your vendors direct questions. Does the AI act on its own or only suggest? Is your data used for training? For example, in LawMOR, Anthropic's Claude suggests matters and drafts time-entry narratives, but it never bills, never sets billed time, never moves trust money, and firm data is never used to train models. You should get answers that clear from every tool you use.

This post is general information, not legal advice; check your state's rules and the final text of SB 574.

Sources