# Ninth Circuit AI Sanctions: Point of Signing

> The Ninth Circuit issued the first precedential federal appeals ruling sanctioning attorneys for AI-fabricated citations. What the point of signing requires.

[Home](https://rankshieldlegal.com/) / [Blog](https://rankshieldlegal.com/blog/) / Litigation & Evidence Citation integrity
# The First Federal Appeals Court AI Sanctions Precedent: What "Responsibility at the Point of Signing" Means
On June 3, 2026, the U.S. Court of Appeals for the Ninth Circuit issued a published, precedential opinion sanctioning two attorneys for AI-fabricated citations and a lack of candor in an immigration matter. As reported by Bloomberg Law and Law360, the consequences included monetary sanctions of roughly $2,500 each, six-month suspensions from practice before the court, and a standing obligation to disclose AI use and personally verify every citation in future filings. The reasoning behind it is not new: responsibility for what a brief says attaches when the lawyer signs and files it, not when a model drafts it.

By [Jamie Kloncz](https://rankshieldlegal.com/about/), Founder, RankShield ** 18 min read ** Published July 15, 2026

The Ninth Circuit did not sanction two lawyers because they used artificial intelligence. It sanctioned them because they filed authorities they had not verified and then failed to be candid about it. In LNU v. Blanche, No. 24-4790, decided June 3, 2026, the court issued what reputable coverage describes as the first precedential ruling from a federal appeals court addressing AI-fabricated citations. The holding a filing attorney should take from it is narrow and old: the duty to confirm that your cited authorities exist, say what you claim, and remain good law is non-delegable, and it is enforced at the point of signature.
This article explains what the court actually decided, why an appellate and precedential ruling carries more weight than the earlier trial-court sanctions, and what the decision asks of anyone who signs a brief. It is written from the perspective of a verification vendor and a security engineer, not a law firm, and it is informational rather than legal advice. Where a specific figure or term appears, it is drawn from the published opinion and its coverage in outlets including Bloomberg Law, Law360, and the Volokh Conspiracy, and it is attributed accordingly.

## What the Ninth Circuit actually decided in LNU v. Blanche
In LNU v. Blanche (June 3, 2026), the Ninth Circuit sanctioned two attorneys over AI-fabricated citations and a lack of candor in an immigration appeal. Per Bloomberg Law and Law360, the order included roughly $2,500 each, six-month suspensions, and future AI-disclosure and personal-verification requirements.
The case arose from an immigration matter. According to the published opinion and its coverage, the attorneys submitted briefs that cited judicial decisions which did not exist and included fabricated quotations attributed to real cases. When the discrepancies surfaced, the panel focused not only on the fabrications themselves but on the failure to disclose that generative AI had produced them. The court treated the combination of unverified citations and a lack of candor as the sanctionable conduct.
As reported by Bloomberg Law and Law360, the panel imposed monetary sanctions of roughly $2,500 on each attorney, suspended both from practice before the Ninth Circuit for six months, and directed them to notify their clients, opposing counsel, and the presiding judge of the order. Coverage in the Volokh Conspiracy notes the court also referred the matter to the State Bar of California and other licensing authorities. These specifics are drawn from the opinion and reputable reporting rather than independently reconstructed here.
The framing the court chose is the part worth memorizing. The problem was not that a tool was used in the drafting process. The problem was that authorities entered a signed filing without the signing attorney confirming they were real and accurately represented. That is a verification failure, and the court measured it against duties that predate generative AI entirely.

## Why an appellate, precedential ruling changes the weight of the risk
Sanctions for AI-fabricated citations are not new. The widely reported landmark is Mata v. Avianca, decided in the Southern District of New York in June 2023, where ChatGPT invented six nonexistent decisions and the attorneys were sanctioned $5,000 under Rule 11 [[7]](#ref-7). In the years since, trial courts across many jurisdictions have issued similar orders. Public trackers, including the Damien Charlotin AI Hallucination Cases database, catalog well over a thousand filings worldwide involving AI-fabricated or misused citations as of mid-2026, and the count should be read as approximate and growing [[6]](#ref-6).
What distinguishes LNU v. Blanche is its posture. Earlier sanctions came from trial courts and functioned as cautionary examples with limited formal reach beyond the parties. A published opinion from a federal court of appeals is precedent within that circuit. It is the kind of authority other courts cite and other lawyers are expected to know. Coverage describes it as the first federal appellate, precedential AI-sanctions ruling, which is why it reads less like an isolated embarrassment and more like a stated standard of conduct [[2]](#ref-2) [[4]](#ref-4).
For a filing attorney, the practical consequence is that the expectation is no longer buried in scattered trial-court orders. It sits in a precedential opinion that opposing counsel, disciplinary bodies, and future panels can point to. That does not create a new duty. It makes an existing one harder to claim ignorance of.
Dimension Earlier trial-court sanctions LNU v. Blanche (9th Cir., 2026)
Court level Federal and state trial courts Federal court of appeals
Precedential weight Persuasive at most; bound to the parties Published, precedential within the circuit [2]
Reference point Mata v. Avianca, $5,000 (S.D.N.Y. 2023) [7] Roughly $2,500 each plus six-month suspensions, per coverage [4]
Forward obligation Usually none beyond the order Standing AI-disclosure and personal-verification duty [3]

Source: Ninth Circuit, No. 24-4790 (2026) Download SVG

## "Responsibility at the point of signing" is the holding that matters
The organizing idea of the opinion is where responsibility attaches. A brief passes through many hands and tools: an associate researches, a database returns results, a generative model drafts, a paralegal formats. None of that transfers accountability. Accountability lands on the person whose signature appears on the filing. As the coverage frames it, the rules are not violated at the point of research and drafting; they are violated at the point of signing and filing.
Reporting on the opinion quotes the court's emphasis that the signature was the attorney's alone and stood as an attestation that he had personally reviewed the brief's contents [[4]](#ref-4). That is the mechanism in one sentence. Signing is not a formality at the end of a workflow. It is the moment the lawyer certifies to the court that the work meets the required standard, and it is the moment the duty is either satisfied or breached.
This is why "the AI generated it" does not function as a defense. The court's logic does not ask which tool produced the error. It asks whether the signer confirmed the filing before certifying it. A confident, fluent fabrication that reaches a court under a lawyer's signature is a failure of that confirmation, regardless of how the text was produced.
Point of signing where the Ninth Circuit locates the duty: not research or drafting, but signature and filing [4]

## The mechanism is existing Rule 11 and the court's inherent authority
It is worth being precise about the legal machinery, because it is not novel. The duties the court enforced flow from Federal Rule of Civil Procedure 11 and from courts' inherent authority to manage the conduct of the lawyers before them. Rule 11 provides that by signing a filing, an attorney certifies that a reasonable inquiry was made and that the legal contentions are warranted. Inherent authority lets a court sanction conduct that undermines the integrity of its proceedings, including a lack of candor.
No new rule was required for this outcome, and none should be assumed from it. There is no confirmed amendment creating a special AI certification duty at work here. The court applied standards that already governed reliance on associates, paralegals, and commercial research services and extended the same reasoning to generative output. The duty of reasonable inquiry has always meant the signer answers for what enters the record. Generative AI is simply the newest source a lawyer might lean on without checking.
Reading it this way removes the temptation to treat the case as an AI-specific curiosity. It is a candor-and-diligence case with a modern fact pattern. The tools changed. The obligation did not.
The mechanism here is existing Rule 11 and inherent authority. This piece does not assert that any new federal rule on AI citations has been adopted; separate proposals exist but remain pending and are outside the scope of this ruling.

## The disclosure and certification requirement the court imposed going forward
Beyond the immediate sanctions, the court attached a forward-looking obligation, which is one of the more instructive parts of the order for other practitioners. As reported, future filings by the attorneys involved must state whether generative AI was used, identify the specific tool, and certify that the signing attorney personally reviewed every citation and quotation [[3]](#ref-3). Coverage indicates this certification is to be made under penalty of perjury.
That remedy is telling because it names the exact gap the sanctions punished. The court did not prohibit AI. It required disclosure of its use and a personal attestation of verification. In effect, the order converts the general duty of reasonable inquiry into an explicit, filing-by-filing checkpoint for these attorneys: confirm existence, confirm accuracy, and say so on the record.
Other lawyers are not bound by this particular order, but the shape of it is a preview of where the standard of care is heading. A growing number of trial judges already require similar AI-use certifications by standing order. A precedential appellate opinion adopting the same logic gives that trend a firmer footing.

## What "personally verified" asks of the signing attorney
Personal verification is not re-researching the brief. It is confirming, for each authority, that it exists, that it is quoted accurately, and that it remains good law. Reported reliability figures around 17 to 33 percent for leading legal AI tools in 2024 are why that check has to run against the actual source [[1]](#ref-1).
The phrase personally verified can sound like a demand to redo all research from scratch. It is not. Reasonable inquiry does not ask the signer to re-derive every proposition of law. It asks the signer to take the steps a competent lawyer would take to confirm three things about each cited authority.
First, that the authority exists and is correctly identified. A fabricated case fails here before any question of what it holds. Second, that the quotation and characterization are accurate, meaning the brief says what the case actually says rather than what a tool asserted it said. Third, that the authority is still good law and has not been overruled, reversed, or superseded. Coverage of the opinion notes the court expected more than confirming a citation exists; a diligent attorney must also read and reason through the authority itself [[4]](#ref-4).
The opinion reportedly cited empirical work on tool reliability, noting research that found leading legal AI research products returned inaccurate answers on a meaningful share of queries, with figures reported around 17 percent for one product and 33 percent for another in 2024 testing [[1]](#ref-1). That error rate is the practical reason personal verification is not optional. The tools present inaccurate output with the same fluency as accurate output, so the fabrication cannot be caught by tone or confidence. It has to be checked against the source.

## A pre-filing citation verification workflow
The reliable way to satisfy the duty the Ninth Circuit described is to make verification a defined step between drafting and signature, applied to every filing rather than left to whoever proofreads under deadline pressure. The steps below are not exotic. They are the checks reasonable inquiry has always implied, run consistently and early enough to matter. RankShield's [AI legal citation checker](https://rankshieldlegal.com/ai-legal-citation-checker/) runs the first existence check for free, and full [citation certification](https://rankshieldlegal.com/citation-certification/) covers the rest.

- **Extract every cited authority** Pull each case, statute, and quotation out of the draft into a list so nothing is verified by skimming. A citation that is never isolated is a citation that is easy to miss.
- **Confirm each authority exists** Check that every cited decision is real and correctly identified against live case-law. A fabricated case fails at this first step, which is exactly where the Lnu sanctions began.
- **Confirm the quotation and characterization** Verify that the brief quotes and describes each authority accurately, rather than repeating what a model asserted the case held. This is the second failure the court named.
- **Confirm it is still good law** Check that the authority has not been overruled, reversed, or superseded, so the argument does not rest on a decision that no longer stands.
- **Record who verified and how** Keep a per-filing record of what was checked, by whom, and against what source, so the personal review the court expects is documented rather than assumed.
- **Gate the signature on the result** Treat verification as a precondition for filing, not a courtesy pass. A flagged citation stops the brief before it reaches the court and before a signature certifies it.

## Why individual carefulness is not the same as a control
A firm can tell every lawyer to check every citation and still be exposed. Carefulness is a habit, and habits degrade under exactly the conditions that produce fabricated filings: a compressed deadline, a delegated draft, a confident block of text that reads like every other block of text. Relying on each attorney to independently catch a fluent fabrication is relying on the least distracted person on the worst day to notice something designed to look correct.
A control is different from a habit. A control is a repeatable step that runs on every filing regardless of who drafted it or how tight the timeline was, and it produces a record that the step ran. The Ninth Circuit's forward-looking remedy points in this direction. By requiring a per-filing certification that citations were personally reviewed, the court effectively told these attorneys to convert a habit into a checkpoint. Other firms can adopt the same posture voluntarily, before a court requires it of them.
This is the governance point behind the doctrine. The duty is non-delegable, which means it cannot be handed off. But the mechanics of satisfying it can be systematized, so that the signer's certification rests on a defined process rather than on memory and good intentions under pressure. Related reading covers the [mechanics of AI hallucinations in legal filings](https://rankshieldlegal.com/ai-hallucination-legal-filings/) and the [escalating sanctions for fabricated AI citations](https://rankshieldlegal.com/fake-ai-citation-sanctions/).

## Where certification before signature fits, honestly
RankShield Legal is a verification vendor, not a law firm. Its flagship capability is built for the exact gate the Ninth Circuit described: before a filing is signed, resolve each cited authority against live case-law to check that it exists, that it is quoted accurately, and that it remains good law, then record the result. That turns the sentence "we checked the citations" into a verifiable record of what was checked and when.
Being precise about the claim matters here. RankShield does not claim to make AI drafting safe, and it does not claim to be "hallucination-free," because no tool can honestly make that claim. What it does is certify which cited authorities are real and accurately represented, and flag the ones that are not, so a fabricated case is caught at the point of prevention rather than discovered in a sanctions order. The free citation checker performs the existence check; certification extends to quotation accuracy and good-law status.
A verification layer does not lift the non-delegable duty off the signing attorney, and nothing sold by a vendor can. The lawyer still certifies the filing and still answers for it. What the layer offers is evidence that the reasonable inquiry actually happened: a per-filing record aligned with the personal-verification certification the Ninth Circuit now expects. The value is narrow and concrete, which is to catch the fabricated citation before a judge does and to be able to show the work.

## The practical takeaway for anyone who signs a brief
Read against the profession's older rules, LNU v. Blanche does not announce anything philosophically new. It restates that the signer owns the filing and applies that to generative output in a precedential opinion that is now hard to overlook. The behavior it punishes is the failure to verify and the failure to be candid, not the use of a tool.
For a filing attorney, three concrete lessons follow. Verify every cited authority against the actual source before signing, because existence, accuracy, and good-law status are all in scope. Disclose AI use where a court, standing order, or the emerging standard of care calls for it, and be candid the moment a problem surfaces rather than after a panel finds it. And treat verification as a system that runs on every filing, so the certification you make at signature is backed by a record rather than by hope. This is informational content from a security and engineering perspective, not legal advice, and it is not a substitute for the independent judgment of licensed counsel.
Not legal advice. This article is informational and reflects the perspective of a verification vendor and a security engineer, not an attorney. It does not create an attorney-client relationship. Specific figures and terms are attributed to the published opinion and reputable coverage; consult the primary sources and licensed counsel before relying on any detail.

Test yourself
## Test yourself on the point of signing
Four questions on what LNU v. Blanche asks of anyone who signs a brief.

- 1 Where does the Ninth Circuit locate the duty to verify cited authorities? At research and drafting At the point of signing and filing With the AI vendor that produced the draft **Answer:** At the point of signing and filing The opinion locates responsibility at signature and filing. Signing certifies the attorney personally reviewed the brief, so the duty is satisfied or breached there.
- 2 What legal mechanism did the court apply? A new AI-specific certification rule Existing Rule 11 and the court's inherent authority A binding state bar advisory opinion **Answer:** Existing Rule 11 and the court's inherent authority No new rule was needed. The court applied Rule 11's reasonable-inquiry duty and the court's inherent authority, extending standards that already governed reliance on associates and research services.
- 3 What does personal verification require for each cited authority? Re-researching the entire brief from scratch Confirming it exists, is quoted accurately, and remains good law Confirming only that the citation format is correct **Answer:** Confirming it exists, is quoted accurately, and remains good law Reasonable inquiry means confirming existence, accurate quotation, and good-law status for each authority, not re-deriving every proposition of law.
- 4 What error rates did the opinion cite for leading legal AI research tools in 2024 testing? Around 1 to 2 percent Around 17 to 33 percent Around 50 to 60 percent **Answer:** Around 17 to 33 percent The opinion referenced research reporting inaccurate answers on a meaningful share of queries, with figures around 17 percent for one product and 33 percent for another.
Honest self-check. There is no sign-up, and nothing is stored.

Questions answered
## Straight answers to the common questions
The questions readers ask about this topic, answered directly. **No forms, no sales pitch.**

JAMIE KLONCZ · SEO AGENCY NAPLES ************** ONLINE
Pick a question on the left, or search above. You will get the direct answer, the way an answer engine would give it.

← PREV NEXT → [REQUEST ACCESS →](https://rankshieldlegal.com/contact/)

- **What did the Ninth Circuit hold in LNU v. Blanche?** In a published, precedential opinion issued June 3, 2026, in LNU v. Blanche, No. 24-4790, the Ninth Circuit sanctioned two attorneys over AI-fabricated citations and a lack of candor in an immigration matter. As reported by Bloomberg Law and Law360, the order included monetary sanctions of roughly $2,500 each, six-month suspensions from practice before the court, and a requirement to disclose AI use and personally verify citations in future filings. The organizing principle is that responsibility for a filing attaches at the point of signing and filing, not at research or drafting. The court did not prohibit AI use. It punished filing unverified authorities and failing to be candid about how they were produced. These specifics are drawn from the opinion and its coverage rather than independently reconstructed here.
- **Is this the first federal appeals court ruling on AI hallucinations?** Reputable coverage describes LNU v. Blanche as the first precedential ruling from a federal court of appeals sanctioning attorneys for AI-fabricated citations. Earlier sanctions, including the widely reported Mata v. Avianca decision in 2023, came from trial courts and functioned as cautionary examples rather than binding circuit precedent. Public trackers such as the Charlotin AI Hallucination Cases database catalog well over a thousand filings worldwide involving AI-fabricated or misused citations as of mid-2026, and that figure is approximate and growing. What sets the Ninth Circuit ruling apart is its posture: a published appellate opinion is precedent within the circuit, which other courts cite and other lawyers are expected to know. It does not create a new duty so much as make an existing one harder to overlook.
- **Does this ruling create a new Rule 11 requirement for AI?** No. The court applied existing law: Federal Rule of Civil Procedure 11, which provides that signing a filing certifies a reasonable inquiry was made, together with courts' inherent authority to sanction conduct that undermines the integrity of proceedings, including a lack of candor. No new federal rule was needed for this outcome, and none should be assumed from it. The duty to verify cited authorities is the same standard that has long governed reliance on associates, paralegals, and commercial research services, now extended to generative output. Separate proposals to amend Rule 11 specifically for AI citations have been discussed elsewhere, but they remain pending and are not what this ruling turned on. The mechanism here is the existing rule and inherent authority.
- **Why is "the AI generated it" not a defense?** Because the duty of reasonable inquiry rests on the attorney who signs the filing, not on the tool that produced the draft. Under Rule 11, a signature certifies that the signer confirmed the filing, and that responsibility cannot be delegated to a generative model any more than it can be delegated to an associate or a research service. As the Ninth Circuit framed it, the signature is an attestation that the attorney personally reviewed the brief's contents. The court's logic does not ask which tool produced an error. It asks whether the signer verified the work before certifying it. Because leading legal AI tools return inaccurate output on a meaningful share of queries, with figures reported around 17 to 33 percent in 2024 testing, unverified reliance is treated as a failure of diligence rather than an excusable accident.
- **What does "personally verified" require of a lawyer?** It does not require re-researching the entire brief. It requires confirming three things about each cited authority. First, that the authority exists and is correctly identified, since a fabricated case fails before any question of what it holds. Second, that the quotation and characterization are accurate, meaning the brief reflects what the case actually says. Third, that the authority remains good law and has not been overruled, reversed, or superseded. Coverage of the opinion indicates the court expected more than confirming a citation exists; a diligent attorney must also read and reason through the authority itself. The practical way to meet this consistently is to make verification a defined step between drafting and signature that runs on every filing, rather than an ad hoc check performed under deadline pressure.
- **How can a firm build this into its process without banning AI?** Treat verification as a control rather than a habit. A control is a repeatable step that runs on every filing regardless of who drafted it or how tight the timeline was, and it produces a record that the step ran. In practice that means extracting every cited authority, checking each against live case-law for existence, quotation accuracy, and good-law status, recording who verified what, and gating the signature on the result. Tools can assist with the existence and accuracy checks, but they do not lift the non-delegable duty off the signing attorney. RankShield is a verification vendor, and its citation checker performs the existence check with certification extending to accuracy and good-law status. The goal is narrow: catch a fabricated citation before a judge does, and be able to show the verification was actually performed.

## References

- Stanford RegLab / HAI (Magesh, Surani, Dahl, Suzgun, Manning, Ho). Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (finding leading tools returned inaccurate answers on a meaningful share of queries in 2024 testing). 2024. [https://hai.stanford.edu/news/ai-trial-legal-models-hallucinate-1-out-6-or-more-benchmarking-queries](https://hai.stanford.edu/news/ai-trial-legal-models-hallucinate-1-out-6-or-more-benchmarking-queries)
- U.S. Court of Appeals for the Ninth Circuit. LNU v. Blanche, No. 24-4790 (published opinion). Primary source; specific figures in this article are attributed to reputable coverage. June 3, 2026. [https://cdn.ca9.uscourts.gov/datastore/opinions/2026/06/03/24-4790.pdf](https://cdn.ca9.uscourts.gov/datastore/opinions/2026/06/03/24-4790.pdf)
- Law Office of Sabrina Damast, Inc.. Ninth Circuit Sanctions Attorneys for AI-Hallucinated Immigration Briefs (disclosure and personal-verification requirement). June 12, 2026. [https://www.sabrinadamast.com/journal/2026/6/12/ninth-circuit-sanctions-attorneys-for-ai-hallucinated-immigration-briefs](https://www.sabrinadamast.com/journal/2026/6/12/ninth-circuit-sanctions-attorneys-for-ai-hallucinated-immigration-briefs)
- Bloomberg Law. Ninth Circuit Warns of AI Hallucinated Briefs in Sanctions Order (sanction amount, six-month suspension, tool reliability figures). June 2026. [https://news.bloomberglaw.com/litigation/ninth-circuit-warns-of-ai-hallucinated-briefs-in-sanctions-order](https://news.bloomberglaw.com/litigation/ninth-circuit-warns-of-ai-hallucinated-briefs-in-sanctions-order)
- Reason / The Volokh Conspiracy (Eugene Volokh). Ninth Circuit on AI Hallucinations (holding on signature as personal attestation; State Bar referral). June 3, 2026. [https://reason.com/volokh/2026/06/03/ninth-circuit-on-ai-hallucinations/](https://reason.com/volokh/2026/06/03/ninth-circuit-on-ai-hallucinations/)
- Charlotin, D.. AI Hallucination Cases database (well over a thousand filings worldwide as of mid-2026; figure approximate and updated). 2026. [https://www.damiencharlotin.com/hallucinations/](https://www.damiencharlotin.com/hallucinations/)
- U.S. District Court, S.D.N.Y.. Mata v. Avianca, Inc., 678 F. Supp. 3d 443 ($5,000 Rule 11 sanction for six fabricated ChatGPT citations). June 22, 2023. [https://law.justia.com/cases/federal/district-courts/new-york/nysdce/1:2022cv01461/575368/54/](https://law.justia.com/cases/federal/district-courts/new-york/nysdce/1:2022cv01461/575368/54/)

Written by
## [Jamie Kloncz](https://rankshieldlegal.com/about/)
Founder, RankShield
Jamie Kloncz is the founder of RankShield, the verifiable AI and quantum security platform behind RankShield Legal. An engineer by training, he built RankShield after his own devices and business were attacked, including an AI voice-cloning scam that targeted his family, on one conviction: unverifiable security is the real danger, so every consequential action should leave a receipt anyone can independently check.
[More about Jamie →](https://rankshieldlegal.com/about/)

Try it · Free
## Check a citation against live case-law
Paste a citation from an AI-drafted brief and see whether the case actually exists, resolved against live case-law. Free, no sign-up. Then request early access to certify a full filing.
[Try the citation checker](https://rankshieldlegal.com/ai-legal-citation-checker/)

Keep reading
## Related guides
[Litigation & Evidence Federal judges are using AI: what bench-side adoption means for your filings Read guide →](https://rankshieldlegal.com/blog/federal-judges-using-ai-what-it-means/)[Litigation & Evidence New York's Part 161 Makes AI Verification a Court-Wide Duty Read guide →](https://rankshieldlegal.com/blog/new-york-part-161-ai-court-rule/)[Litigation & Evidence AI restrictions in protective orders: the new e-discovery battleground Read guide →](https://rankshieldlegal.com/blog/ai-restrictions-protective-orders-ediscovery/)