California SB 574 Passed Without a Single No Vote. Unless the Governor Vetoes It by September 30, Lawyers Get Their First AI Statute on January 1.
Executive Summary
- California SB 574 passed the Assembly 75 to 0 on August 31, 2026, the Senate concurred the same day with no votes against, and the enrolled bill was presented to Governor Newsom on September 9, 2026. Because it passed before the September 1 cutoff in the California Constitution, the Governor must return it with a veto on or before September 30, 2026 or it becomes a statute without his signature.
- If it becomes law, it takes effect January 1, 2027 and adds Section 6068.1 to the Business and Professions Code. An attorney may not delegate the practice of law to generative AI, may not enter confidential or personal identifying information into a generative AI system that is not access restricted, must take reasonable steps to verify and correct AI output including every case and statutory citation, and must disclose generative AI use to the court for every document submitted.
- It also amends Code of Civil Procedure Section 128.7 so that no brief, pleading, motion or other paper filed in any California court may contain a citation the responsible attorney has not personally verified. Existing Section 128.7 already makes a law firm jointly responsible for violations by its partners, associates and employees, so the exposure runs to the firm, not just the drafter.
- The bill creates no new penalty. Enforcement runs through Section 128.7 sanctions and State Bar discipline. Under existing law a court must report any sanction of $1,000 or more to the State Bar, and the attorney must self report at the same line. A Los Angeles judge’s $999.99 sanction against a State Farm defense lawyer, reported September 14, 2026, sits one cent below it.
- California courts have not waited for the statute. Published appellate opinions in September 2025 and July 2026 imposed $10,000 and $1,500 sanctions and forwarded both to the State Bar, a live tracking database listed 2,041 hallucination decisions worldwide as of September 14, 2026, and the State Bar’s own proposed rule amendments still await Board of Trustees and Supreme Court approval. SB 574 would be the only one of these instruments that is a statute.
In this article
- What SB 574 does and exactly where it stands today
- Why September 30 is the date that matters
- The four duties in new Section 6068.1
- The Section 128.7 amendment: personal verification of every citation
- The $1,000 line between a sanction and a State Bar file
- What California courts have already been doing
- How SB 574 fits with the State Bar guidance and the proposed rules
- Arbitrators, mediators and ADR providers
- What changes for a solo or small firm on January 1, 2027
- A 60 day plan before the effective date
1. What SB 574 does and exactly where it stands today
California SB 574 would be the first statewide statute written specifically to govern how attorneys use generative AI, a description its author and most coverage share and which we found no enacted state statute to contradict. It bars an attorney from delegating the practice of law to generative AI, restricts what client information may go into an AI system, requires verification and correction of AI output including every citation, requires disclosure of AI use to the court, and amends the sanctions statute so no filed paper may contain a citation the attorney has not personally verified. If enacted, it takes effect January 1, 2027.
The bill has already cleared the Legislature. The Assembly passed it 75 to 0 on August 31, 2026, the Senate concurred in the Assembly amendments the same day with no votes against, and the enrolled version was produced on September 4 and presented to the Governor at 2 p.m. on September 9, 2026. You can read the whole thing in the enrolled text on the Legislature’s site, and the tallies are in the roll call record. The bill history showed no action by the Governor as of September 17, 2026.
The author is Senator Tom Umberg, who chairs the Senate Judiciary Committee, with Assembly Member Macedo as coauthor. The Senate first passed the bill 39 to 0 on January 29, 2026, and no committee in either house recorded a no vote. That matters for the veto calculus. A bill with zero recorded opposition in either house is a hard one to return.
Everything in this article is drawn from the enrolled text of September 4, 2026, not from the introduced version. That distinction has tripped up coverage of other California bills this session, and the enrolled SB 574 is narrower in places than the summaries circulating. In particular, the bill creates no standalone penalty, no filing certification form, and no private right of action. What it creates is a set of statutory duties that plug into two enforcement systems that already exist: Code of Civil Procedure Section 128.7 sanctions and State Bar discipline.
Two scope points before the detail. First, the attorney provisions in new Business and Professions Code Section 6068.1 apply to every California licensee who uses generative AI to assist in the practice of law, with no firm size threshold. The State Bar’s own demographics page listed 201,143 active attorneys as of September 16, 2026. Second, the citation rule in Section 128.7 applies to any paper filed in any court, so it reaches a two page ex parte application in a limited civil case exactly as it reaches an appellate brief.
2. Why September 30 is the date that matters
September 30, 2026 is the last day the Governor can veto SB 574, and if he does nothing it becomes a statute anyway. That is not a political prediction. It is the text of Article IV, Section 10(b)(2) of the California Constitution: any bill passed by the Legislature before September 1 of the second year of the session and in the Governor’s possession on or after September 1 “that is not returned on or before September 30 of that year becomes a statute.”
SB 574 passed both houses on August 31 and reached the Governor on September 9, so it sits squarely inside that rule. There are three possible outcomes. He signs it and it becomes law. He does nothing and it becomes law. He returns it with a veto message on or before September 30. Two of the three end in the same place, and the third requires an affirmative act against a bill that did not draw a single no vote in either chamber.
The effective date follows from a different constitutional provision. Article IV, Section 8(c)(1) provides that a statute enacted at a regular session goes into effect on the January 1 following a 90 day period from the date of enactment. SB 574 contains no urgency clause and no delayed operative date. Enactment in September 2026 therefore means an effective date of January 1, 2027, which gives a firm roughly three months from the Governor’s deadline to be ready.
This is the same constitutional clock we walked through for employers in our SB 947 article, where the Legislature passed the No Robo Bosses Act one day before the cutoff. The difference is the vote. SB 947 cleared the Assembly 53 to 14. SB 574 cleared it 75 to 0.
3. The four duties in new Section 6068.1
Section 1 of SB 574 adds Section 6068.1 to the Business and Professions Code, and it imposes one prohibition and three affirmative duties on any attorney who uses generative AI to assist in the practice of law. It opens with a savings clause at subdivision (a)(1) that nothing in the section abrogates the existing duty of competence and diligence, which signals that the Legislature intends these duties to add to, not replace, the Rules of Professional Conduct.
The prohibition: no delegation of the practice of law
Subdivision (a)(2) is one sentence: “An attorney shall not delegate the practice of law to generative artificial intelligence.” The bill does not define “delegate,” and it does not define “the practice of law” beyond the existing body of California case law on that phrase. Read with the rest of the section, the plain meaning is that a lawyer may use the tool to assist but may not let it substitute for the lawyer’s own judgment, research and review. That tracks the State Bar’s position that “the lawyer’s duty of candor to the tribunal cannot be delegated to AI,” and it is close to the language courts have already used when sanctioning lawyers who filed AI drafts they never read.
Duty one: confidentiality by system design
Subdivision (a)(3)(A) bars an attorney from entering “confidential, personal identifying, and other nonpublic information” into a generative AI system unless access to what the attorney inputs “is restricted to the attorney and persons authorized by the attorney under obligations to protect the confidentiality of the information.” The test is about the system, not the prompt. A consumer tier service whose provider can access, retain, review or train on prompts is unlikely to meet it, and an enterprise deployment with contractual confidentiality, access controls and no training on customer inputs can, but neither label is decisive. The determination turns on the specific plan, contract, retention settings, access controls and data use terms in place for that account.
The definition of personal identifying information at subdivision (b)(2) says the term “includes all the following” and then lists nine categories, so the list is illustrative rather than exhaustive: driver’s license numbers, dates of birth, Social Security numbers, National Crime Information and Criminal Identification and Information numbers, addresses and phone numbers of parties, victims, witnesses and court personnel, medical or psychiatric information, financial information, account numbers, and anything sealed by court order or deemed confidential by rule or statute. That list is lifted almost word for word from California Rules of Court, rule 10.430(d)(1), which the Judicial Council adopted effective September 1, 2025 to govern generative AI use by court staff. The Legislature took the rule the courts wrote for themselves and applied it to the bar.
Duty two: verify and correct
Subdivision (a)(3)(B) requires the attorney to 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.” Two features of that language matter. The standard is reasonable steps, not strict liability. And the correction duty attaches to “any material used by the attorney,” which is broader than court filings. A demand letter, a client memo or a contract draft is material used by the attorney.
Duty three: disclose to the court
Subdivision (a)(3)(C) requires the attorney to “disclose the use of generative artificial intelligence to the court for all documents submitted to the court,” and separately to “consider whether to disclose” AI use when it is used to create content provided to the public. The court disclosure is mandatory and applies to all documents. The public content disclosure is a duty to consider, not a duty to disclose. The bill prescribes no form, no location and no wording for the court disclosure, and the only instruction it gives the Judicial Council, in new Code of Civil Procedure Section 180, is to publicly revisit Standard 10.80 of the Standards of Judicial Administration, which governs judges’ own use of generative AI. Because the bill prescribes no form or placement, implementation may develop through Judicial Council action, statewide rules, local court rules, standing orders and case by case practice, and firms should expect the format to settle over time.
GOVERNANCE INSIGHT
SB 574 regulates the lawyer’s process, not the tool
Nothing in the bill names a product, bans a category of software or requires a vendor certification. Every duty is written as something the attorney does: chooses a system with restricted access, verifies, corrects, discloses. That is the same design choice the NIST AI RMF and ISO/IEC 42001 make at the organizational level. It means a firm cannot buy compliance by switching tools. It has to be able to show what its people did with the output, and that is a records question before it is a technology question.
4. The Section 128.7 amendment: personal verification of every citation
Section 3 of SB 574 amends Code of Civil Procedure Section 128.7 to add a new paragraph (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.” That is the provision with teeth, because Section 128.7 is California’s counterpart to Federal Rule 11 and already carries a sanctions mechanism. Note that the two standards are different. Section 6068.1 requires reasonable steps to verify AI output generally. Section 128.7(b)(2) is narrower and stricter: it applies only to citations in filed papers, and there the standard is personal verification by the responsible attorney. Some summaries render the phrase as “personally read and verified.” The enrolled text says “personally verified,” and that is the language quoted here.
Three things about the amendment deserve a close read. First, the verb is “personally.” The duty cannot be discharged by a paralegal, a citator, or a second AI tool. That codifies what the Court of Appeal held in July 2026 in Del Biaggio v. Bansen, discussed below, where a lawyer’s plan to have his paralegal check AI generated citations “would have been inappropriate even if it had not gone awry.”
Second, the duty attaches to “an attorney responsible for submitting the pleading.” It does not say every attorney whose name appears on the caption. A senior judge of the D.C. Court of Appeals wrote separately this month to make exactly that distinction, noting that requiring every listed signatory to personally check every citation “would be impractical, unduly burdensome, and unnecessarily duplicative.” California’s text avoids that problem by tying the duty to responsibility for submission, but a firm should decide in writing who that person is on each filing. Naming that attorney does not relieve other signing or supervising lawyers of their own duties under Rules 3.3, 5.1 and 5.3 or under the general Section 128.7 certification that every signer makes.
Third, the rest of Section 128.7 is unchanged, and the unchanged parts are what make the amendment expensive. Subdivision (c)(1) provides that “absent exceptional circumstances, a law firm shall be held jointly responsible for violations committed by its partners, associates, and employees.” Subdivision (c) also preserves the 21 day safe harbor. Under (c)(1) a party’s sanctions motion may not be filed if the challenged paper is withdrawn or corrected within 21 days of service, and under (c)(2) a court acting on its own motion issues an order to show cause that likewise falls away if the paper is withdrawn or corrected within 21 days of service. Subdivision (d) limits any sanction to “what is sufficient to deter repetition,” and allows nonmonetary directives, a penalty paid into court, or, on motion, payment of the moving party’s fees and expenses. Subdivision (d)(1) provides that monetary sanctions may not be awarded against a represented party for a violation of paragraph (2) of subdivision (b), which after the renumbering is the citation verification paragraph. On the plain text, and subject to how courts read the renumbering, that points the monetary exposure for an unverified citation at the attorney and the firm rather than the client. No court has construed the new paragraph yet.
| Provision | What SB 574 adds | What already existed | Enforced through |
|---|---|---|---|
| Bus. and Prof. Code 6068.1(a)(2) | No delegation of the practice of law to generative AI | Duty of competence, Rule 1.1; State Bar guidance that candor cannot be delegated to AI | State Bar discipline |
| 6068.1(a)(3)(A) | No confidential, personal identifying or nonpublic information into a system that is not access restricted | Bus. and Prof. Code 6068(e)(1) duty to preserve client secrets; Rule 1.6 | State Bar discipline |
| 6068.1(a)(3)(B) | Reasonable steps to verify all output including every citation, and to correct hallucinated output in any material used | Rule 3.3 candor; Rule 1.1 competence; 2023 and 2026 State Bar Practical Guidance | State Bar discipline |
| 6068.1(a)(3)(C) | Mandatory disclosure of generative AI use to the court for all documents; consider disclosure for public content | No statewide disclosure duty for attorneys | State Bar discipline; court authority over filings |
| Code Civ. Proc. 128.7(b)(2) | No filed paper may contain a citation the responsible attorney has not personally verified | 128.7(b) certification after “an inquiry reasonable under the circumstances”; firm joint responsibility at (c)(1); 21 day safe harbor | Court sanctions on motion or on the court’s own motion |
| Code Civ. Proc. 1282.1 | Arbitrator may not delegate any part of decisionmaking to generative AI; must disclose reliance on AI generated material outside the record | Ethics Standards for Neutral Arbitrators; duty of neutrality | Existing Arbitration Act disclosure and vacatur grounds as informed by the new duty; complaints under the State Bar’s voluntary ADR certification program |
5. The $1,000 line between a sanction and a State Bar file
SB 574 creates no penalty of its own, so the practical consequence of a violation is whatever a court does under Section 128.7 and whatever the State Bar does with the referral, and the referral is triggered by existing law at exactly $1,000. Business and Professions Code Section 6086.7(a)(3) requires a court to notify the State Bar of “the imposition of any judicial sanctions against an attorney, except sanctions for failure to make discovery or monetary sanctions of less than one thousand dollars ($1,000).” Section 6068(o)(3) puts the same duty on the attorney, who must self report any sanction at or above that line within 30 days. Under Section 6086.7(c), the State Bar “shall investigate any matter reported.”
That threshold is the reason a $999.99 figure drew so much attention in mid September. Los Angeles Superior Court Judge Elizabeth Bradley ordered a senior associate at Musick, Peeler and Garrett, defending State Farm in a homeowner coverage case over fire damage at a house in Carson, to pay a $999.99 sanction after seven case citations in her pretrial filings turned out not to exist. According to the ABA Journal’s account of the associate’s filing, she had used an AI legal research tool she believed was connected to the firm’s Westlaw subscription and believed it ran its own citation check. The order, reported by Reuters and Law360 on September 14, 2026, sits one cent below the reporting threshold. Both Section 6086.7(a)(3) and Section 6068(o)(3) exclude monetary sanctions of less than $1,000, so on that amount neither the court’s reporting duty nor the attorney’s self reporting duty is triggered. Nothing in the public reporting says the court chose the figure for that reason, and this article does not suggest it did.
Treat that as a data point about where the line sits, not as a safe harbor, and note that a report to the State Bar opens a review rather than establishing misconduct. In the two published California appellate opinions on AI citations, discussed next, the courts imposed $10,000 and $1,500 and expressly ordered the clerk to forward the opinion to the State Bar under Section 6086.7(a)(3). Once SB 574 is in force, a court weighing a sanction will be applying a statute that says the citation duty is personal and nondelegable, and the argument that the lawyer did not know the tool could fabricate authority is gone.
6. What California courts have already been doing
California appellate courts have already imposed the personal verification duty that SB 574 would codify, in two published opinions that any small firm lawyer should read before January. Both were certified for publication precisely so they would bind and warn the rest of the bar.
The first is Noland v. Land of the Free, L.P. (2025) 114 Cal.App.5th 426, filed September 12, 2025 by the Second District. The appellant’s opening brief contained 23 case quotations, and 21 of them were fabrications. Counsel admitted at oral argument that he wrote initial drafts, “enhanced” them with ChatGPT, ran the enhanced briefs through other AI platforms to check for errors, and did not read them before filing. The court imposed what it called a conservative sanction of $10,000, ordered the opinion forwarded to the State Bar, and required counsel to give a copy to his client and certify that he had done so. Its statement that “it is a fundamental duty of attorneys to read the legal authorities they cite in appellate briefs or any other court filings” is the line the First District quoted back ten months later.
The second is Del Biaggio v. Bansen, A174647, filed July 10, 2026 by the First District. The lawyer had used generative AI to add case law to an appellate brief during trial recesses in another matter and intended to have his paralegal check the citations before filing. The brief ended up attributing fictitious quotations to real California Supreme Court and Court of Appeal decisions. The court held that the protocol “would have been inappropriate even if it had not gone awry,” because the State Bar’s guidance says a lawyer must review all outputs, and it found that waiting more than three months after learning of the problem to withdraw the misstatements, only days before oral argument, violated the duty of candor under Rule 3.3(a)(1). The sanction was $1,500 payable to the court, with the same State Bar referral and client notification orders as Noland. The Court of Appeal noted that People v. Alvarez (2025) 114 Cal.App.5th 1115 had imposed the same $1,500 figure for a fictitious quotation, a nonexistent case and two off point citations.
The pattern is national. On August 28, 2026 a federal judge in Atlanta fined plaintiff’s counsel $8,000 in Booker v. The Kroger Co., calculated as $1,000 for each of four fake, false or misleading authorities and then doubled because counsel had told the court under questioning that he “did not use artificial intelligence” before conceding in writing that he had. On September 3, 2026 the D.C. Court of Appeals struck an entire brief filed for a Deutsche Bank trust after four cited cases turned out not to exist, refused the firm’s request to simply redact them as “effectively no sanction at all,” and referred the matter to disciplinary counsel. As local reporting noted, that left the client without a written argument in an active foreclosure appeal.
The volume is the part that should register with a firm owner. Damien Charlotin’s AI Hallucination Cases database, a live tracker that courts themselves have cited, listed 2,041 decisions worldwide as of its September 14, 2026 update. When we published our state bar ethics map five weeks earlier, the count was about 1,800. Treat both numbers as snapshots. The database is updated continuously, adds older matters as they surface, and covers pro se litigants and other participants as well as lawyers, so the difference is not a clean count of new sanctions against attorneys.
| Matter | Court and date | What went wrong | Outcome |
|---|---|---|---|
| Noland v. Land of the Free | Cal. Ct. App., 2d Dist., September 12, 2025 (published) | 21 of 23 quotations in the opening brief fabricated; briefs not read before filing | $10,000 to the court; opinion forwarded to State Bar; client notification ordered |
| Del Biaggio v. Bansen | Cal. Ct. App., 1st Dist., July 10, 2026 (published) | Fictitious quotations attributed to real cases; verification left to a paralegal; correction delayed three months | $1,500 to the court; opinion forwarded to State Bar; client notification ordered |
| Booker v. The Kroger Co. | N.D. Ga., August 28, 2026 | Four fake, false or misleading authorities; counsel denied AI use in open court, then admitted it in writing | $8,000 to the court registry ($1,000 per authority, doubled); proof of ethics CLE required |
| Douglas v. Deutsche Bank National Trust Co. | D.C. Ct. App., September 3, 2026 (published order) | Four nonexistent cases from a general purpose AI search tool; attorney did not verify | Entire brief stricken; referral to Office of Disciplinary Counsel |
| Meni-Siliga v. State Farm | L.A. Superior Court, order reported September 14, 2026 | Seven nonexistent citations in motions in limine from an AI research tool believed to be connected to Westlaw | $999.99, one cent under the $1,000 State Bar reporting threshold |
7. How SB 574 fits with the State Bar guidance and the proposed rules
SB 574 would be the third layer of California AI regulation for lawyers, sitting on top of nonbinding State Bar guidance that has existed since November 2023 and proposed Rules of Professional Conduct amendments that are still in the comment process. Understanding which layer binds you today, and which will bind you on January 1, is the whole compliance question.
The first layer is the Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law, issued by the State Bar’s Standing Committee on Professional Responsibility and Conduct in November 2023 and replaced by a 2026 version approved by the Board of Trustees on May 14, 2026. The 2026 version was prepared at the California Supreme Court’s request and adds coverage of agentic AI tools, which it defines as systems that can plan and execute tasks without human prompting. It is guidance, not a rule, but it is the document the Court of Appeal measured the lawyer’s verification protocol against in Del Biaggio, and it states plainly that “a lawyer must review all outputs produced using AI tools for accuracy.”
The second layer is the proposed rule amendments. On August 22, 2025 the California Supreme Court sent the State Bar a letter directing it to consider incorporating the guidance into the binding Rules of Professional Conduct. On March 13, 2026 the committee approved proposed amendments to Rules 1.1, 1.4, 1.6, 3.3, 5.1 and 5.3 for a 45 day comment period that closed May 4, 2026. The committee revised the proposals and circulated them for a second comment period that closed August 6, 2026. The amendments still require Board of Trustees adoption and Supreme Court approval, and as of this writing neither has happened. We covered the substance in the ethics map. The short version is that they would make independent verification of AI output, an expanded definition of “reveal” for confidentiality purposes, and firm level AI policies express disciplinary obligations.
SB 574 is the third layer, and it differs from the other two in three ways. It is a statute, so it does not depend on the Supreme Court’s rulemaking calendar. It reaches into the Code of Civil Procedure, so a trial judge can enforce the citation rule through sanctions without any disciplinary proceeding. And it contains a mandatory court disclosure duty that neither the guidance nor the proposed rules impose. The guidance requires a lawyer to consider client disclosure and the proposed Rule 1.4 comment would require client communication when AI use materially affects the representation. Only SB 574 requires telling the court.
| Instrument | Status as of September 17, 2026 | Binding? | Court disclosure duty? |
|---|---|---|---|
| State Bar Practical Guidance (2026 version) | Approved by Board of Trustees May 14, 2026; replaces November 2023 version | No, but cited by courts as persuasive authority on existing duties | No; requires lawyer to consider client disclosure |
| Proposed amendments to Rules 1.1, 1.4, 1.6, 3.3, 5.1, 5.3 | Approved for comment March 13, 2026; comment periods closed May 4 and August 6, 2026; awaiting Board and Supreme Court action | Not yet | No; client communication when use materially affects representation |
| Rules of Court, rule 10.430 and Standard 10.80 | In effect since September 1, 2025 | Yes, for court staff and judicial officers only | Not applicable to attorneys |
| SB 574 | Enrolled and presented to the Governor September 9, 2026; veto deadline September 30; effective January 1, 2027 if enacted | Yes, if enacted, as statute | Yes, for all documents submitted to the court |
Getting your AI governance in order
Dynamic Comply helps small and medium sized businesses build a real AI governance foundation using the NIST AI RMF and ISO/IEC 42001.
8. Arbitrators, mediators and ADR providers
SB 574 also regulates arbitrators directly and rewrites the complaint process for State Bar certified ADR providers, which matters to any small firm whose lawyers sit as neutrals or whose clients are routinely sent to private arbitration. Section 5 adds Code of Civil Procedure Section 1282.1, and Section 2 amends Business and Professions Code Section 6173.
New Section 1282.1(b)(1) provides that “an arbitrator shall not delegate any part of their decisionmaking process to any generative artificial intelligence tool.” Subdivision (b)(2) adds that an arbitrator “shall not rely on information generated by generative artificial intelligence outside the record without making appropriate disclosures to the parties beforehand and, as far as practical, allowing the parties to comment on its use.” Like the attorney section, it opens with a savings clause preserving the duties of neutrality, competence and diligence, and it uses the same definition of generative AI. For a party, the practical effect is a statutory basis to demand disclosure. Undisclosed reliance on AI generated material outside the record could also support arguments about arbitral fairness or misconduct under the existing statutory grounds for vacating an award, but the bill does not create a new vacatur ground and review of awards remains highly deferential.
The Section 6173 amendment is procedural but consequential, and it reaches only participants in the State Bar’s voluntary certification program for ADR firms, providers and practitioners, not every arbitrator or mediator in California. SB 574 would require the State Bar or an approved professional organization to receive, investigate and resolve complaints that a certified provider failed to comply with applicable ethical standards, would require certified providers and licensees to produce relevant records on request in that process, and would make the proceedings private and confidential, with a Public Records Act exemption and a legislative finding to support it. After a decision, the State Bar may publicly disclose the name of a provider found to have violated a standard and the action taken. The compelled production duty in subdivision (b)(2)(B) applies only to a mediation where one of the parties is self represented, and privileged material still requires written consent from every holder of the privilege.
9. What changes for a solo or small firm on January 1, 2027
For a solo or small firm, SB 574 would change four things on January 1, 2027 if it is signed or allowed to become law: which AI tools may receive client information, who signs off on citations, what goes on every court filing, and what the firm can prove afterward. None of the four requires new technology. All four require decisions that most small firms have not yet written down.
Tool selection becomes a confidentiality determination
The Section 6068.1(a)(3)(A) test is whether access to what you input is “restricted to the attorney and persons authorized by the attorney under obligations to protect the confidentiality of the information.” That is a question about the vendor’s terms and configuration. A firm should be able to point to the contract clause or enterprise setting that restricts access and disables training on its inputs for each tool it permits, and should treat any tool it cannot document that way as off limits for anything in the nine category definition of personal identifying information. The 2026 Practical Guidance says reasonable efforts “require more than reliance on generalized marketing assurances,” and that is the right standard to apply here as well.
One named attorney owns every citation on every filing
Because Section 128.7(b)(2) ties the duty to “an attorney responsible for submitting the pleading,” the firm should name that attorney on each filing and that attorney should personally open every cited authority in a primary source before signing. A paralegal cite check is still valuable as a first pass, but after Del Biaggio it does not discharge the duty, and after SB 574 it will not satisfy the statute. For a firm that files in federal court as well, this is the same verification a Rule 11 inquiry already demands, so one protocol can serve both.
A disclosure line on every document submitted to the court
Section 6068.1(a)(3)(C) is written as an absolute: disclosure “for all documents submitted to the court.” Until the Judicial Council or individual courts prescribe a format, the conservative reading is a short, factual statement in or with each filing stating that generative AI was used in its preparation and that the responsible attorney has reviewed and verified the content and citations. A firm that has decided not to use generative AI on court documents at all avoids the disclosure question, but it should then be able to show that its people actually followed that policy, because the D.C. case involved a firm whose stated policy prohibited AI drafting.
Records that show the verification happened
Every enforcement path under SB 574 runs through a court or the State Bar asking what the attorney did. A one page verification log per filing, listing each authority, the primary source consulted and the date and initials of the attorney who checked it, is the cheapest evidence a small firm can generate. It is also what turns a reasonable steps standard from an argument into a record. The same log answers the State Bar’s proposed Rule 5.1 comment on internal AI policies if that amendment is eventually adopted.
10. A 60 day plan before the effective date
A small firm can be ready for SB 574 in about 60 days with four steps, and it should start after September 30 whether the bill is signed or simply left to become law. This is a defensible baseline, not a guarantee, and it is deliberately built so that nothing is wasted if the Governor vetoes, because every step also answers the State Bar guidance and the published sanctions opinions that already apply.
Days 1 to 15: inventory and classify. List every generative AI tool anyone in the firm uses for client work, including features embedded in research platforms, document management systems and email. For each, record the vendor, the plan tier, whether inputs are used for training, who can access prompts and outputs, and whether the contract contains a confidentiality obligation. Sort them into three buckets: cleared for confidential information, cleared for nonconfidential work only, and prohibited. Most firms find at least one consumer tool in the first bucket that belongs in the third.
Days 16 to 30: write the policy. A two to three page policy is enough. It should name the cleared tools, state the nine categories of personal identifying information from Section 6068.1(b)(2) as a hard stop for anything outside the first bucket, assign one responsible attorney per filing for citation verification, describe the verification log, and set the court disclosure language the firm will use. Have every attorney and staff member sign it. Rules 5.1 and 5.3 already make supervising lawyers responsible for staff use, and the proposed comments would make firm policies express.
Days 31 to 45: run the protocol on live filings. Apply the verification log to every filing for two weeks and time it. Firms that do this find the cost is measured in minutes per authority, which is small next to a $1,500 to $10,000 sanction, a State Bar investigation and a client notification order. Fix whatever slows it down, usually access to a primary source database for every attorney rather than a shared login.
Days 46 to 60: close the loop. Brief the firm on the effective date and on what the disclosure line will look like. Check the local rules of the courts you file in most for any AI specific requirement, because those will fill the gap the statute leaves on format. Calendar a review for mid 2027, when the Judicial Council’s revisit of Standard 10.80 and the State Bar’s rule amendments may both have landed. If the firm has lawyers who sit as arbitrators, add the Section 1282.1 disclosure duty to their engagement templates.
The theme across all four steps is the one in the Governance Insight above. SB 574 does not ask whether you used AI. It asks what you did next, and whether you can show it.
Talk to Ross about your AI governance needs
Every business has different AI governance requirements. Let us talk about yours.
Frequently Asked Questions
Is California SB 574 law yet?
Not as of September 17, 2026. The Legislature passed SB 574 on August 31, 2026 with no votes against in either house and presented it to the Governor on September 9, 2026. Under Article IV, Section 10(b)(2) of the California Constitution, the Governor must return it with a veto on or before September 30, 2026 or it becomes a statute without his signature.
When would SB 574 take effect?
January 1, 2027. The bill has no urgency clause and no delayed operative date, so under Article IV, Section 8(c)(1) of the California Constitution a statute enacted in September 2026 goes into effect on the January 1 following a 90 day period from enactment.
Does SB 574 ban lawyers from using ChatGPT or other generative AI?
No. It bars delegating the practice of law to generative AI and bars entering confidential, personal identifying or other nonpublic information into a generative AI system unless access to those inputs is restricted to the attorney and persons under confidentiality obligations. A lawyer may use any tool for work that does not involve that information, and may use an access restricted tool for confidential work, provided every output is verified and corrected and AI use is disclosed to the court.
What are the penalties for violating SB 574?
SB 574 creates no penalty of its own. A citation violation is enforced through Code of Civil Procedure Section 128.7, which allows a court to impose sanctions sufficient to deter repetition, including a penalty paid into court or the opposing party’s fees, and which holds a law firm jointly responsible absent exceptional circumstances. Any sanction of $1,000 or more must be reported to the State Bar under Business and Professions Code Sections 6068(o)(3) and 6086.7(a)(3), which opens a State Bar review but is not itself a finding of misconduct.
Do I have to tell the court every time I use AI on a filing?
Yes, if SB 574 becomes law. New Business and Professions Code Section 6068.1(a)(3)(C) requires an attorney who uses generative AI to assist in the practice of law to disclose that use to the court for all documents submitted to the court. The bill prescribes no form or wording, and the only direction it gives the Judicial Council is to revisit Standard 10.80 on judges’ own AI use, so the format will be settled by local practice.
Can a paralegal verify AI generated citations under SB 574?
Not as a substitute for the attorney. Amended Code of Civil Procedure Section 128.7(b)(2) provides that no filed paper may contain a citation that an attorney responsible for submitting the pleading has not personally verified. The Court of Appeal reached the same result in Del Biaggio v. Bansen on July 10, 2026, holding that a plan to have a paralegal check AI generated citations was inappropriate and imposing a $1,500 sanction with a State Bar referral.
About the author
Ross J. is the founder of Dynamic Comply, an AI governance, compliance, and cybersecurity consulting firm based in Leesburg, Virginia. He brings more than 15 years of federal cybersecurity experience across the Department of State, the Department of Defense, and the Department of Homeland Security, and holds the CGRC certification along with credentials as a GSDC AI Compliance Lead Implementer and Auditor and Certified Ethical Hacker.
Related articles
This article is provided for general informational purposes and reflects the state of the law as of September 2026. It is not legal advice. Regulations in this area are changing quickly. Confirm current requirements and consult qualified counsel before making decisions for your organization.