SB 574 amends California Code of Civil Procedure section 128.7 by adding a specific prohibition: attorneys must not file briefs, pleadings, motions, or other court papers containing citations that the attorney has not personally read and verified, and this prohibition explicitly covers citations supplied by generative artificial intelligence. The bill also supplies a statutory definition of “generative artificial intelligence.”
This change inserts a concrete verification duty into the existing certification-and-sanctions regime. For litigators, in-house counsel, and legal-tech vendors, the practical effect is immediate: firms must adopt workflows and documentation practices to prove that a human reviewed every authority cited, or face sanctions under the existing 128.7 framework (subject to the section’s monetary-sanction limits and procedural safeguards).
At a Glance
What It Does
The bill adds paragraph (b)(2) to Section 128.7, forbidding any filed court paper from containing citations that the submitting attorney has not personally read and verified, and it defines "generative artificial intelligence." It leaves the existing certification and sanctions framework intact, including the 21‑day safe-harbor for withdrawing or correcting challenged papers.
Who It Affects
Practicing litigators and law firms across California, legal-research and generative‑AI vendors whose outputs are used in litigation, and courts that will evaluate alleged violations under Section 128.7. It also affects clients who rely on faster, AI‑assisted drafting workflows.
Why It Matters
The statute converts a soft ethical risk—relying on unverified AI outputs—into a fixed procedural duty enforceable by sanctions. Firms must change vetting, training, and vendor-contract practices or face responsibility for AI-generated errors.
More articles like this one.
A weekly email with all the latest developments on this topic.
What This Bill Actually Does
Section 128.7 has long required attorneys to certify that filings are made for proper purposes and that legal and factual contentions have a reasonable basis; it also authorizes courts to sanction violations. SB 574 builds on that framework by adding a targeted rule about citations.
The new paragraph requires that any brief, pleading, motion, or similar paper filed in court must not include citations the responsible attorney has not personally read and verified. The provision expressly calls out citations produced by "generative artificial intelligence," which the bill defines broadly as systems that can generate synthetic text, images, video, or audio resembling their training data.
Practically, the bill does two things at once. First, it narrows the gap created when attorneys rely on third‑party outputs—especially AI—to assemble authorities, by insisting on personal verification rather than blind reliance.
Second, it retains the existing 128.7 enforcement tools: adversaries can move for sanctions (with the statutory safe‑harbor to withdraw or correct within 21 days), and courts can issue show‑cause orders or impose appropriate sanctions subject to the section’s limits. Notably, the text prohibits monetary sanctions against a represented party for violations of the new paragraph (b)(2), but preserves other monetary and nonmonetary remedies in many circumstances.For law firms, the operational implications are concrete.
Firms should adopt affirmative verification procedures—documented checks that an attorney actually opened and read each cited source and confirmed its accuracy and relevance, including pin cites, quotations, and procedural posture. Contract terms with legal‑AI vendors will need to address liability and representation of accuracy.
Courts and opposing counsel will likely ask for proof of verification in show‑cause proceedings, so preserving metadata, search logs, or signed verifier attestations will be a likely compliance strategy.The bill also creates some unresolved edges. The definition of generative AI is broad and multimedia‑oriented even though the prohibition targets citations, which are typically textual and legal; the statute excludes disclosures and discovery requests from its reach; and the title mentions arbitrators while the body applies to filings in court.
Those mismatches will leave judges and practitioners to sort the statute’s scope and practical enforcement questions when disputes arise.
The Five Things You Need to Know
SB 574 adds Section 128.7(b)(2): a filed brief, pleading, motion, or similar paper “shall not contain any citations that the attorney responsible for submitting the pleading has not personally read and verified,” explicitly including citations provided by generative AI.
The bill defines “generative artificial intelligence” as systems that generate derived synthetic content—including text, images, video, and audio—based on training data.
A motion for sanctions under Section 128.7 remains subject to the statute’s 21‑day safe‑harbor: the challenged paper must be withdrawn or corrected within 21 days of service (or another period set by the court) before the motion is presented to the court.
The statute bars monetary sanctions against a represented party specifically for violations of paragraph (2), though other sanctions and fee awards remain available to deter misconduct.
Section 128.7 does not apply to disclosures and discovery requests, responses, objections, or motions—those materials are explicitly excluded from the verification requirement.
Section-by-Section Breakdown
Every bill we cover gets an analysis of its key sections.
Signature and contact requirement for filings
Subdivision (a) preserves the longstanding rule that every pleading or similar paper must be signed by an attorney of record (or a self‑represented party), and that an unsigned paper is subject to being stricken unless the omission is promptly corrected. That baseline signature rule anchors personal responsibility; SB 574 leaves it in place so the new verification duty ties back to the signer’s certification obligation.
Existing certification duties remain
Subdivision (b)(1) restates the lawyer’s existing certification obligations—no improper purpose, legal contentions warranted by law or nonfrivolous argument, factual allegations supported or likely to be after investigation, and warranted denials. SB 574 does not replace those duties; instead it supplements them with the citation verification rule, so a violation can be framed both as a breach of the longstanding certification and as falling within the new paragraph (b)(2).
New rule: personal reading and verification of citations; AI defined
This is the operative addition. It requires that any citation in a court filing must have been personally read and verified by the submitting attorney; the rule explicitly covers citations produced by generative AI. The bill supplies a broad, technology‑neutral definition of generative AI that includes multiple media types—text, images, video, audio—which will matter when courts parse whether a given tool fits the definition. Practitioners must interpret “read and verified” in context; the text gives no procedural checklist, leaving room for judicial development.
Sanctions procedure and motion timing, including safe‑harbor
Subdivision (c) preserves the motion and show‑cause mechanisms: a party seeking sanctions must bring a separate motion describing the alleged conduct, serve it under Section 1010, and allow the 21‑day safe‑harbor for withdrawal or correction before filing it with the court. The court may also act on its own motion by issuing an order to show cause and giving the target 21 days to correct. The provision continues the familiar rule that firms can be held jointly responsible for violations by their partners, associates, or employees, subject to considerations of due diligence.
Limits on sanctions and monetary awards
Subdivision (d) imposes proportionality limits: sanctions must be limited to what deters repetition and may include nonmonetary directives, penalties paid into court, or, in some cases, payment of reasonable attorney fees incurred as a direct result. Crucially, the statute bars monetary sanctions against a represented party specifically for violations of paragraph (b)(2), and it restricts monetary sanctions on the court’s own motion unless the court issued its show‑cause order before a voluntary dismissal or settlement.
Discovery exclusion and application to pleadings
Subdivision (g) excludes disclosures and discovery requests, responses, objections, and motions from the section’s scope—so the citation verification rule does not reach ordinary discovery materials. Subdivision (i) reiterates the section’s applicability to complaints or petitions filed on or after January 1, 1995, and to subsequent pleadings in those matters; that carry‑forward language is a quirk of the statute that preserves its reach across pending matters and should be noted when assessing which cases are covered.
This bill is one of many.
Codify tracks hundreds of bills on Justice across all five countries.
Explore Justice in Codify Search →Who Benefits and Who Bears the Cost
Every bill creates winners and losers. Here's who stands to gain and who bears the cost.
Who Benefits
- Opposing parties and clients who face fewer AI‑hallucination or mis-cited authorities—fewer baseless or mistaken citations can reduce time spent rebutting incorrect legal authorities.
- Courts, which gain a statutory tool to hold attorneys accountable for relying on unreliable AI outputs and to reduce the volume of filings that misquote or misstate authorities.
- Legal‑research and verification vendors that build tools to document human verification and provide audit trails, creating a market for compliance solutions.
- Clients who prioritize accuracy and risk control, particularly institutional litigants who will benefit from clearer documentation that counsel actually verified cited authorities.
Who Bears the Cost
- Law firms and individual attorneys, who must allocate attorney hours to verify citations manually, revise workflows, and maintain proof of verification—small firms and solos are likely to feel this burden most acutely.
- Generative‑AI vendors, which may need to add disclaimers, change product features, or provide logging and provenance tools to support verification and contract language to limit liability.
- Courts and judges, which may see an uptick in 128.7 motions or show‑cause proceedings as counsel tests the new rule and opposing parties weaponize verification disputes.
- Clients who face higher legal bills as firms shift time from drafting to verification and compliance tasks, especially in high‑volume or rapid‑turnaround matters.
Key Issues
The Core Tension
The central dilemma is straightforward: protect the integrity of court records and deter AI‑driven inaccuracies by imposing a human verification duty, versus imposing a compliance burden that slows advocacy, raises costs (especially for small practices), and risks chilling useful AI adoption—without clear standards for what verification requires or how courts should evaluate proof of verification.
SB 574 pushes an important accuracy norm into a sanctionable procedural duty, but it leaves numerous implementation details unresolved. The phrase “personally read and verified” lacks statutory content: the bill does not say whether verification requires checking primary sources, confirming procedural posture, validating pin cites, or documenting the steps taken.
That uncertainty invites litigation over what counts as reasonable verification in different practice contexts—criminal vs. civil, routine motions vs. complex appellate briefs.
The bill’s definition of generative AI is broad and media‑agnostic, which risks overbreadth when applied to citation verification (a textual task). Tools that use AI components for vendor research, indexing, or citation‑formatting might fall within the definition even if they do not produce the argumentative text of a brief.
The discovery exclusion narrows scope, but does not resolve questions about using AI to summarize depositions or to generate arguments based on discovery materials that are later cited in filings. Finally, the statutory ban on monetary sanctions against a represented party for violations of paragraph (b)(2) limits the teeth of enforcement and may channel disputes into nonmonetary sanctions or fee awards to movants—a choice that reshapes deterrence incentives.
Implementation will depend heavily on courts developing standards and on firms adopting recordkeeping practices that can survive judicial scrutiny. Expect doctrinal skirmishes over whether metadata, search logs, attestation checklists, or a verifier’s signed declaration suffice as proof.
The title’s reference to arbitrators, without corresponding text, also creates uncertainty about whether and how the rule should be applied in private arbitration contexts—an uncertainty parties will likely test in future proceedings.
Try it yourself.
Ask a question in plain English, or pick a topic below. Results in seconds.