Key takeaways
- In Noland v. Land of the Free (2025), a California Court of Appeal sanctioned counsel $10,000 after 21 of 23 brief quotations proved fabricated.
- SB 574, effective January 1, 2027, bars filing papers with citations the responsible attorney has not personally verified.
- SB 574 also requires disclosing generative AI use to the court and limits entering confidential information into AI systems.
- A written cite-check log, kept with the matter, is the simplest way to show verification happened.
- Firm sign-in, blocked personal AI accounts and retained prompt logs make the verification trail provable.
Why is verifying AI citations a filing issue now?
Verifying AI citations is now a filing issue because California courts are sanctioning attorneys for fabricated citations, and starting January 1, 2027, state law makes personal verification of every citation part of what a filing attorney certifies. AI tools produce hallucinated citations in three forms: cases that do not exist, real cases with invented quotations and real cases cited for points they never made. A citation an AI tool supplied is treated the same as one an associate typed in by hand. If it is wrong, the signer owns it.
The leading California case is Noland v. Land of the Free, L.P. (2025, B331918), the first published California appellate opinion on AI-fabricated citations. The Court of Appeal found that 21 of the 23 case quotations in the appellant's opening brief were fabricated, with more false citations in the reply. It sanctioned appellant's counsel $10,000, directed the clerk to send the opinion to the State Bar, and published the opinion as a warning that no brief, motion or pleading should contain citations counsel has not personally read and verified.
The court also declined to award fees to the other side, in part because respondent's counsel never flagged the fake citations. The lesson cuts both ways: check what your opponent cites, too.
What does SB 574 require of California attorneys?
SB 574, signed September 30, 2026 as Chapter 858 of the Statutes of 2026, takes effect January 1, 2027. It amends Code of Civil Procedure section 128.7 so that a filed paper may not contain a citation the responsible attorney has not personally verified, and it adds Business and Professions Code section 6068.1 on attorney use of generative AI.
As enacted, section 6068.1 includes these duties:
- An attorney may not delegate the practice of law to generative AI.
- An attorney must take reasonable steps to verify the accuracy of AI output, including case and statutory citations, and correct erroneous or hallucinated output in any material the attorney uses.
- An attorney may not enter confidential, personally identifying or other nonpublic information into a generative AI system unless access is restricted to the attorney and persons the attorney authorizes who are obligated to keep it confidential.
- An attorney must disclose the use of generative AI to the court for all documents submitted to the court, and consider whether to disclose it when AI is used to create content provided to the public. The statute does not prescribe the form of that disclosure.
Citation violations run through the existing section 128.7 sanctions process, which holds a law firm jointly responsible for violations by its partners, associates and employees absent exceptional circumstances. The statute does not define "personally verified," so read it conservatively and ask counsel how it applies to your practice.
What do the ethics rules already say about AI output?
The ethics guidance already expected verification before SB 574. The State Bar of California's Practical Guidance for the Use of Generative Artificial Intelligence warns that AI output can be false or inaccurate and says a lawyer must critically review, validate and correct it. ABA Formal Opinion 512 (July 29, 2024) says uncritical reliance on AI output can produce misleading representations to courts and that the level of independent review depends on the task.
Federal court works the same way. Federal Rule of Civil Procedure 11(b)(2) makes the signer certify that legal contentions are warranted by existing law or a nonfrivolous argument for changing it. A case that does not exist cannot support either.
How do you verify an AI citation, step by step?
Verify each authority against the original source, never against the AI tool's own summary or link. Work from a written cite list so nothing is skipped.
- Extract every authority from the draft into a cite list: cases, statutes, rules of court, regulations and secondary sources.
- Pull each one from an official or trusted source, such as the official reporter, the court's opinion page, the legislature's code site or your research platform. Do not click through a link the AI tool generated.
- Confirm the case exists as cited: party names, court, year, volume and page all match.
- Read the pinpoint page and confirm it supports the proposition in your brief. Real cases cited for the wrong point are the harder error to catch.
- Match every quotation word for word, including ellipses and brackets.
- Run a citator for subsequent history: reversed, depublished, review granted or superseded.
- Check that the authority can be cited at all. California Rules of Court, rule 8.1115 generally bars citing unpublished California appellate opinions.
- For statutes and rules, confirm you are citing the version in effect for your issue.
- Record the check in a cite-check log saved to the matter file: authority, source used, who checked it, date and result.
- Run the same check on the other side's key citations before you rely on or respond to them.
Who at the firm should do the checking?
A paralegal or associate can do the first pass, but the attorney responsible for the filing should personally confirm every citation before it goes out. That reading follows the "personally verified" language in SB 574 and the Noland court's warning. It also fits the supervision duties attorneys already have over the people and tools that help prepare a filing.
In practice, that means the cite-check log travels with the draft. The signing attorney reviews the log, spot-checks any authority marked as AI-sourced and initials it before filing. If nobody can show that log later, the firm has a harder time explaining what happened if a bad citation surfaces.
How can firm IT make verification easier to prove?
Firm systems can make the verification trail automatic instead of relying on memory. The controls that matter most are:
- Approved AI tools, used only through firm sign-in, so prompts and outputs sit in accounts the firm controls.
- Blocked personal AI accounts on firm devices, which supports the confidentiality duty in section 6068.1.
- Prompt and output logs retained with the matter, so the firm can show what the tool produced and what the attorney changed.
- A cite-check log template stored in the document management system, with a required field before a filing is marked final.
- Standard AI disclosure language, approved by litigation leadership, ready for January 1, 2027.
Our article on giving attorneys AI legal research tools without a confidentiality gap covers the vendor terms and sign-in controls in more detail.
For the confidentiality side of section 6068.1, see how to block privileged documents from reaching AI tools.
What should a law firm do before January 1, 2027?
Use the remaining weeks of 2026 to make verification a written, repeatable step. A short list for this quarter:
- Add the ten-step cite check to every filing checklist and calendar it at least a day before each filing deadline.
- Decide who signs off on cite-check logs and how long the firm keeps them.
- Draft and approve the firm's AI disclosure language for court filings.
- Inventory the AI tools attorneys and staff actually use, and move approved work onto firm accounts.
- Train attorneys and staff on SB 574, Noland and the firm's own procedure, and keep an attendance record.
- Confirm that prompt logs and AI output are retained in a way you could produce if a court asks.
How can s90 help a law firm show its AI controls work?
s90 sets up AI platform policy enforcement and prompt monitoring for law firms, so approved tools run under firm sign-in, personal AI accounts are blocked on firm devices and prompt logs are kept where the firm can retrieve them. If you want help turning your verification procedure into controls you can show a court or the State Bar, we are happy to talk it through.
