California: AI Ethics Guidance for Law Firms

Bottom line for a 5-50 attorney California firm: SB 574, signed September 30, 2026, takes effect January 1, 2027. It adds Business and Professions Code section 6068.1, which requires an attorney who uses generative AI to keep nonpublic information out of unrestricted AI systems, take reasonable steps to verify outputs, and “Disclose the use of generative artificial intelligence to the court for all documents submitted to the court.” It also bars filed papers from containing citations the responsible attorney has not personally verified. Until then, COPRAC’s Practical Guidance (2026 revision, Board-approved 2026-05-14) sets out the State Bar’s expectations. Proposed amendments to the Comments to six Rules remain proposed, not adopted: no Board of Trustees action had been published as of 2026-10-01, and adoption would still need Board action and California Supreme Court approval. Firms that cannot document confidentiality vetting, citation verification, and supervision policies are exposed on Rules 1.1, 1.6, 3.3, and 5.1 today, before any new rules take effect.

Start with: a written AI use policy, vendor due-diligence records for each AI tool, and a citation verification log for court filings.

COPRAC Practical Guidance for the Use of Generative AI

Citation: State Bar of California, Committee on Professional Responsibility and Conduct, Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (2026 revision, approved by the State Bar Board of Trustees May 14, 2026, replacing the version approved November 16, 2023). Primary source PDF.

Status: Non-binding practice guidance. Not a numbered formal opinion in the CAL series. The State Bar calls the document a “living document” subject to update without versioning notice. There is no binding bar-issued rule on attorney AI use in California today.

Versioning hazard worth knowing about. The State Bar republishes each revision at the same filename, and the document carries no revision date on its face. A firm that downloaded the guidance in 2025 and a firm that downloads it today get materially different documents from the same URL, with no changelog to compare. Check the substance rather than the link.

What the 2026 revision added

The revision states that it “replaces the 2023 version and, at the request of the California Supreme Court, addresses the unique challenges presented by the use of agentic AI.” That framing matters: the additions are not a general refresh, they are aimed at systems that act rather than merely draft.

Lawyers “must not deploy agentic systems in a manner that allows the system to make substantive legal determinations, communicate legal advice, prepare and file pleadings, or otherwise act in a representative capacity without meaningful lawyer supervision and review.” Separately, lawyers “must not permit AI systems to autonomously file documents, communicate with the court, or make representations on the lawyer’s behalf.”

The billing section also tightened: general-purpose AI subscriptions are treated as overhead the firm absorbs, rather than a cost passed through to clients.

For a firm evaluating agentic tooling, those two sentences are the operative constraint. A workflow where an agent drafts and a lawyer reviews before filing is contemplated; a workflow where an agent files, corresponds with the court, or answers a client without review is not.

What it requires (mandatory language)

• Confidential client information may not be entered into any AI tool without first verifying adequate confidentiality and security protections.
• Before adopting an AI tool, the lawyer must verify its security, confidentiality, and data retention protocols. COPRAC frames this as consulting an IT or cybersecurity expert. Firms without dedicated IT staff (including solos) typically satisfy the verification requirement by documenting a review of the vendor’s SOC 2 / ISO 27001 report, Data Processing Addendum, and privacy policy. The Vendor Due Diligence Checklist walks the same review.
• Each platform’s terms of service must be reviewed for data handling, third-party sharing, and model training policies.
• All AI-generated citations and factual claims require personal verification before use in court filings or client work product.
• Time saved by AI efficiency cannot be billed to clients; only actual time spent refining prompts and reviewing output may be billed.
• Erroneous, hallucinated, biased, or harmful AI-generated content must be corrected before use.

What it recommends (“should” language)

• Disclose AI use when it materially affects scope, cost, or nature of representation.
• Use AI output only as a starting point; add independent legal analysis.
• Avoid AI products that train on input data or share responses with other users.
• Establish firm-level supervision policies for staff and associate AI use.
• Pursue ongoing education about AI capabilities and limitations.

Notable gaps

Several gaps stand out. The guidance does not define which vendors meet the “adequate protections” threshold. No clear trigger appears for when client disclosure is legally required (vs. recommended). AI in intake or marketing contexts goes unaddressed. Cross-border data exposure via cloud-based AI vendors gets no treatment. The State Bar publishes no changelog despite calling the document a living document.

California Judicial Council Standard 10.80 and Rule 10.430

The Judicial Council adopted Rule 10.430 of the California Rules of Court and Standard 10.80 of the Standards of Judicial Administration, both effective September 1, 2025. They differ in force and in whom they address:

• Rule 10.430 is mandatory and binds courts. “Any court that does not prohibit the use of generative AI by court staff or judicial officers must adopt a generative AI use policy by December 15, 2025,” covering court staff’s use for any purpose and judicial officers’ use outside their adjudicative role. Each policy must, among other things, prohibit entering confidential information into public generative AI systems and require reasonable steps to verify generative AI material.
• Standard 10.80 is guidance for judicial officers. It addresses generative AI use “for any task within their adjudicative role” in “should” terms: a judicial officer “Should take reasonable steps to verify that generative AI material, including any material prepared on their behalf by others, is accurate” and “Should consider whether to disclose the use of generative AI if it is used to create content provided to the public.”

Neither binds attorneys. Why they matter for attorneys: SB 574’s new attorney duties in Business and Professions Code section 6068.1 track Standard 10.80’s wording on nonpublic information, verification, and correction of hallucinated output, and SB 574 adds Code of Civil Procedure section 180, directing the Judicial Council to “publicly revisit, and revise as necessary, Standard 10.80” as generative AI develops. Courts also retain CCP § 128.7 sanctioning authority over attorneys who file AI-hallucinated citations.

What federal courts in California require for AI use in filings

No district-wide AI standing order has been adopted in any of California’s four federal districts (N.D. Cal., E.D. Cal., C.D. Cal., S.D. Cal.). Individual judge orders exist in some chambers. Firms should review the assigned judge’s standing orders at the opening of each federal matter. The operative federal default is Fed. R. Civ. P. 11.

What California state courts require for AI use in filings

No statewide court rule requires attorneys to disclose AI use in California state court today. Standard 10.80 and Rule 10.430 (above) govern court-side use, not attorney filings. That changes by statute on January 1, 2027, when SB 574’s Business and Professions Code section 6068.1 requires an attorney who uses generative AI to disclose that use “to the court for all documents submitted to the court,” and its amendment to CCP § 128.7 bars filed papers from containing citations the responsible attorney has not personally verified (see below). Whether any individual California court has adopted an attorney-facing AI disclosure requirement under its Rule 10.430 policy is an open question; firms should check local rules at the trial-court level when filing.

What AI-related rules are pending in California?

What this means today: the COPRAC Practical Guidance is the operative standard of care now and will remain so whether or not the proposed amendments are adopted. Do not wait for the Board of Trustees vote to write a policy or a verification protocol. The proposals amend the Comments to the Rules, not the rule text, and under Rule 1.0(c) Comments “are not a basis for imposing discipline but are intended only to provide guidance for interpreting and practicing in compliance with the rules.”

Proposed amendments to Rules 1.1, 1.4, 1.6, 3.3, 5.1, and 5.3

On March 13, 2026, COPRAC approved proposed amendments for public comment. According to the State Bar’s staff report, the California Supreme Court directed the State Bar on 2025-08-22 to consider whether the principles of the November 2023 Practical Guidance “should be incorporated into comments for the California Rules of Professional Conduct” (Agenda Item 6.3). The first comment period closed 2026-05-04.

The amendments were not adopted at the close of that comment period. COPRAC revised the package in response to the comments received and, at its June 12, 2026 meeting, approved modified proposals for a second 45-day public comment period closing August 6, 2026 at 11:59 p.m. Public comment page. That window has closed. COPRAC’s August 7, 2026 agenda listed a “Discussion of Artificial Intelligence” (agenda), and its September 11, 2026 agenda listed “Discussion of Artificial Intelligence and Possible Request to Circulate Updated Rules of Professional Conduct Proposals for Public Comment or Approval for Submission to the Board of Trustees” (agenda). As of October 1, 2026, no minutes or outcome for that item had been posted, the Board of Trustees’ September 17-18, 2026 agenda contained no AI item (agenda), and the State Bar had published no Board action on the second-round package. Adoption would still require Board action and then approval by the California Supreme Court, so adopted Comment amendments remain some distance off.

The revision narrowed the single most-quoted obligation in the March draft. Proposed Rule 1.1 Comment [2] had read that “a lawyer must independently review, verify, and exercise professional judgment regarding any output generated by the technology.” The June version requires professional judgment over “all aspects of that use, including… the inputs and outputs,” while scoping the mandatory verification duty to cited legal authority. A firm that wrote its verification protocol against the March language drafted to a stricter standard than the one now proposed. The revision also anchors the definition of “artificial intelligence” to Cal. Gov. Code section 11549.64, and Rule 3.3 was carried through unchanged from the first round.

If adopted, the amendments would add the following to the Comments (the State Bar’s staff report notes “the proposed changes are to the comments of the rules, rather than the text of the actual rules”):

• Rule 1.1: New Comment requiring lawyers to exercise professional judgment over the inputs and outputs of AI use, with verification required for cited legal authority.
• Rule 1.4: New Comment requiring disclosure when AI use “presents a significant risk or materially affects the scope, cost, manner, or decision-making process.”
• Rule 1.6: New Comment defining “reveal” to include exposing confidential information to AI systems where exposure creates “material risk.”
• Rule 3.3: New Comment requiring verification that cited authorities are accurate and not “fabricated, misstated, or taken out of context.” The comment explicitly includes AI-generated citations.
• Rule 5.1: Modified Comment requiring “reasonable efforts to establish internal policies and procedures governing the use of AI.”
• Rule 5.3: Updated comment clarifying supervisory obligations extend to AI tools.

SB 574, enacted: generative AI duties for attorneys and arbitrators

Author: Senator Tom Umberg. The bill passed the Senate on January 30, 2026; the Senate concurred in the Assembly amendments (39-0) on August 31, 2026; and it was presented to the Governor on September 9, 2026. The bill history records “Approved by the Governor” and “Chaptered by Secretary of State. Chapter 858, Statutes of 2026” on September 30, 2026 (bill status; text, enrolled version of September 4, 2026, the latest posted as of October 1, 2026). The bill sets no operative date of its own, so under Cal. Const. art. IV, § 8(c)(1) it takes effect January 1, 2027. The signing is also recorded on the timeline.

What the enacted text provides:

• New Bus. & Prof. Code § 6068.1. “An attorney shall not delegate the practice of law to generative artificial intelligence.” An attorney who uses generative AI to assist in the practice of law must not enter confidential, personal identifying, or other nonpublic information into a system whose access is not restricted to the attorney and persons under confidentiality obligations; must take reasonable steps to “Verify the accuracy of generative artificial intelligence outputs, including, but not limited to, the accuracy of all case and statutory citations” and to “Correct any erroneous or hallucinated output in any material used by the attorney”; and must “Disclose the use of generative artificial intelligence to the court for all documents submitted to the court and consider whether to disclose the use of generative artificial intelligence if it is used to create content provided to the public.”
• Amended CCP § 128.7(b)(2). “A brief, pleading, motion, or any other paper filed in any court shall not contain any citations that an attorney responsible for submitting the pleading has not personally verified, including any citation provided by generative artificial intelligence.” Section 128.7’s existing sanctions provisions apply to violations of subdivision (b).
• New CCP § 180. Directs the Judicial Council to “publicly revisit, and revise as necessary, Standard 10.80” to reflect the development of generative AI.
• New CCP § 1282.1. An arbitrator “shall not delegate any part of their decisionmaking process to any generative artificial intelligence tool,” and may not rely on generative AI information outside the record without disclosure to the parties beforehand.

Unlike the proposed Comment amendments, which under Rule 1.0(c) would not be a basis for discipline, section 6068.1 is a statutory duty in the State Bar Act. The enacted text does not say what form the court disclosure takes; firms may wish to watch for court rules or forms implementing it before January 1, 2027.

Enacted California AI legislation: AB 2013 and SB 942

Two California generative AI transparency statutes were signed in September 2024. AB 2013 requires developers of generative AI systems made publicly available to Californians to post documentation about their training data; the posting duty applied on or before January 1, 2026 (Civ. Code § 3111). SB 942 (the California AI Transparency Act) requires providers of generative AI systems that have over 1,000,000 monthly visitors or users and are publicly accessible in California to offer a free AI-detection tool and to embed latent disclosures in AI-generated image, video, and audio content. SB 942 was originally set to take effect January 1, 2026, but AB 853 (Stats. 2025, ch. 674) amended Bus. & Prof. Code § 22757.6 so that it became operative August 2, 2026. Neither statute regulates lawyer-to-client AI use directly, but both feed firm vendor due diligence: a procured AI tool’s AB 2013 training-data disclosure and SB 942 disclosure mechanism are inputs to the COPRAC Rule 1.6 confidentiality analysis. See the California Generative AI Statutes framework page for the full compliance walkthrough.

How do malpractice carriers in California treat AI use?

No California malpractice carrier has published explicit AI-specific underwriting guidance in materials reviewed for this entry. COPRAC guidance sets out the State Bar’s expectations under Rules 1.1, 1.6, and 3.3. From January 1, 2027, SB 574 adds statutory duties in Business and Professions Code section 6068.1; the proposed amendments, by contrast, would change only the Comments to the Rules, which are interpretive guidance and “are not a basis for imposing discipline” (Rule 1.0(c)).

California courts have already issued sanctions for AI-hallucinated citations under existing authority. In Lacey v. State Farm Gen. Ins. Co., No. 2:24-cv-05205-FMO-MAA (C.D. Cal.), Special Master Michael R. Wilner imposed $31,100 in sanctions on May 6, 2025 against two plaintiffs’ firms. The supplemental brief at issue contained multiple fabricated citations and quotations generated with AI tools. CourtListener docket. In United States v. Hayes, No. 2:24-cr-00280 (E.D. Cal. Jan. 17, 2025), 2025 WL 235531, the court sanctioned an Assistant Federal Defender whose motion to unseal cited and quoted a non-existent case, finding that his “inaccurate and misleading statements were not inadvertent as claimed, but knowing and made in bad faith” (order, ECF No. 62; CourtListener docket).

What does my California malpractice carrier ask about AI at renewal?

No California malpractice carrier has published AI-specific application items in materials reviewed for this entry. The COPRAC Practical Guidance is the de facto substantive standard the carrier or broker will reference at renewal. Plan on producing four items: a written AI use policy; vendor due-diligence records with the IT or security review COPRAC explicitly requires; citation-verification logs for court filings; and a disclosure decision log for matters where AI materially affected scope or cost. A disclosure decision log documents the evaluation that proposed Comment [5] to Rule 1.4 would describe: the duty to keep a client reasonably informed “includes evaluation of the lawyer’s communication obligations concerning the lawyer’s use of technology, including artificial intelligence” throughout the representation. Confirm application items directly with your broker.

What documentation should a California firm keep on file?

Month one (foundational)

1. (Owner: managing partner + firm administrator) Written AI use policy covering confidentiality vetting, citation verification, billing practices, and supervision protocols. Required by implication of COPRAC guidance and, if adopted, the proposed Rule 5.1 amendment.
2. (Owner: litigation lead) Citation verification log or checklist for each court filing. The Practical Guidance, the proposed Rule 3.3 amendment, and SB 574’s amendment to CCP § 128.7 all converge on personal verification of every cited authority. From January 1, 2027, section 6068.1 also requires disclosure to the court of generative AI use in documents submitted; firms may wish to record each disclosure alongside the verification log.
3. (Owner: firm administrator) Training records showing attorneys and staff understand AI limitations and firm policy.

Months two and three (operational documentation)

1. (Owner: firm administrator + outside IT) Vendor due diligence records for each AI tool in use: data handling, model training, third-party sharing, and breach posture. Required by COPRAC and, if adopted, the proposed Rule 1.6 amendment.
2. (Owner: outside IT vendor or partner-of-record) Verification record for each AI tool’s security, confidentiality, and retention posture. COPRAC frames this as consulting an IT or cybersecurity expert. For firms without dedicated IT staff, the operative substitute is a documented review of the vendor’s SOC 2 / ISO 27001 report, Data Processing Addendum, and privacy policy filed in the vendor record.
3. (Owner: managing partner) Supervision protocol covering associate and staff AI use under Rules 5.1 and 5.3.

Months four to six (per-matter discipline)

These are recurring practices, not one-time projects: each new matter exercises them.

1. (Owner: matter lead attorney) Disclosure decision log for matters where AI materially affected scope or cost. It documents the evaluation that proposed Comment [5] to Rule 1.4 would describe; firms may wish to start the log now to avoid retrofitting.

We are building practitioner resources for firms working through this list. The newsletter covers new resources as they ship.