AI-fabricated citation confirmed in Bianco ballot case

For anyone searching for a Chad Bianco 650,000 ballots Supreme Court case legal analysis, the first correction is jurisdictional: this is a California Supreme Court matter, not a U.S. Supreme Court case. The second correction is procedural: as of Aug. 26, 2026, the filing issue is a confirmed AI-citation-fabrication risk record, not a reported sanction event.

The ballot dispute itself is large enough to pull national attention. The California Supreme Court framed the case around whether the Riverside County sheriff’s seizure of approximately 650,000 ballots was legal, with the court weighing the matter after the ballots were taken from election officials in connection with an investigation into alleged election irregularities.[1]

The docket posture matters because it controls what can safely be said. The Riverside voter matter, Cervantes v. Bianco, is listed as S295866, and the Attorney General matter, Attorney General v. Bianco, is listed as S295901.[2] The California Supreme Court granted review and stayed the investigation in April 2026 under a pause-and-preserve order, leaving the legality of the seizure for the court to decide.[3] The cases were argued on Aug. 24, 2026, and no decision had issued by Aug. 26, 2026.

That is enough election-law context for this record. The professional-responsibility issue is narrower: a quotation attributed to a real case was challenged as non-existent, and the responding lawyer reportedly acknowledged that AI was used and the citations were not verified before the filing went in.

The citation issue did not begin as an abstract AI concern

The cleanest documentary starting point is the reply brief filed by the UCLA Voting Rights Project on March 30, 2026. Democracy Docket reported that the brief challenged a quotation attributed to Common Cause v. Board of Supervisors, stating that the quotation “appears nowhere in the opinion.” The same filing described “a concerning pattern” of “non-existent quotes and made-up holdings” in the sheriff’s opposition brief.[4]

That sequence is important. A fabricated-citation record is stronger when it does not rest on a lawyer’s after-the-fact apology alone. Here, the first step is adversarial: opposing counsel identified a specific quotation, tied it to a cited case, and asserted in a filed brief that the quoted language was not in the opinion. The second step broadened the point from one bad line to an asserted pattern in the filing. Those are filed-brief assertions, not a judicial finding.

The third step came at oral argument. Democracy Docket reported that Bianco’s attorney, Bradley Hertz, acknowledged that the citations had not been verified before filing and that AI was used, quoting him as saying: “That is what we’re all learning about AI: It cannot be relied on, and lawyers have to look and make sure that what they’re saying is accurate.” The same coverage reported a justice’s response: “What a concept!”[5] Courthouse News also reported the high court’s questioning in the ballot-investigation argument.[6]

Until an official transcript, recording, corrective filing, sanction order, or merits opinion is reviewed, that oral-argument exchange should be described as press-reported. But the risk status does change. Before the argument, the record showed an opponent’s filed allegation that a quotation appeared nowhere in the cited authority. After the argument coverage, the record includes a reported on-record admission that AI was used and that the citations were not checked before filing.

What is confirmed, what is reported, and what is still open

The distinction between “confirmed risk incident” and “sanction event” is not cosmetic. A risk team can cite this record for the proposition that an AI-assisted filing allegedly contained a quotation that could not be found in the cited opinion and that counsel reportedly admitted the citations were not verified. It should not cite the record for the proposition that the California Supreme Court sanctioned counsel, found a violation of a professional rule, or referred anyone to the bar.

The narrower label also keeps the case usable. Courts, clients, insurers, and regulators often need a live-incident category before discipline exists. This record fits that category because the filing defect and verification failure are the operational facts. The sanction outcome remains blank.

Why this maps to competence and candor risk

The obvious professional-responsibility mapping is Model Rule 1.1 competence and Model Rule 3.3 candor toward the tribunal. ABA Formal Opinion 512, issued July 29, 2024, was the ABA’s first formal ethics guidance on generative AI tools and directed lawyers using generative AI to consider their competence and candor obligations, among others.[7]

The competence problem is not that a lawyer touched an AI system. It is that a legal filing allegedly carried quotations and holdings that were not verified against the source materials before submission. A workflow that permits generated legal text to move from draft to filing without source checking exposes the lawyer, the firm, and the client to the same practical failure: the tribunal is asked to rely on authority that the advocate has not actually confirmed.

The candor problem is equally direct. If a brief represents that a case says something it does not say, the issue is no longer limited to research quality. The court receives a legal proposition dressed as authority. Whether the misstatement was intentional, negligent, or generated through an AI workflow is a separate question for any later discipline or court order. For present risk classification, the relevant fact is the unverified representation to the court.

For internal controls, this is the failure mode usually described as a fabricated quotation from a real case. It is different from an invented case name, and different again from a correct citation used for a proposition the case does not support. Those distinctions are useful because each failure is caught at a different point in review. A case-existence check alone would not catch a false quotation from a real opinion; the reviewer has to open the opinion and compare the language.

For a taxonomy of comparable legal-AI failures, see the site’s discussion of invented cases, fabricated quotations, and unsupported propositions in the Knox Kercher legal AI hallucination probe. For the ethics baseline, the related ABA Formal Opinion 512 compliance playbook and 2026 attorney professional-responsibility guide are the better next stops than the election-law docket.

Why the missing sanction still matters

A sanction record answers a different question from this one. Once a court imposes a monetary sanction, strikes a filing, orders notice to a client, or refers a lawyer for discipline, the professional-responsibility analysis has an adjudicated consequence. Here, no such consequence has been reported as of Aug. 26, 2026.

That makes this record useful for contrast with legal-AI matters that have already crossed the enforcement line. The site’s Ninth Circuit AI-accountability sanctions record and Akerlund v. Atlas Air rebuke record are better examples when the question is what courts have done after non-existent or defective citations reached them. The Bianco ballot case is earlier in the lifecycle: the risk event is visible, but the consequence has not arrived.

That earlier posture is exactly why a knowledge-management lawyer should not wait for a sanction order to log it. The control lesson is already available. A quotation from a real authority has to be checked against the opinion, and every cited proposition has to survive a source-level review before filing. A protocol that would have caught this incident is the same one described in the site’s AI legal-practice verification workflow and ChatGPT ethics guide for lawyers.

The election-law background should stay in the background

The broader Riverside dispute includes claims about vote discrepancies that state and election-law sources have treated as unfounded. The Brennan Center summarized the background by describing a claimed 45,896-vote discrepancy and the registrar’s rebuttal that the actual discrepancy was 103 votes, with variance rates far below the 2% recount threshold cited in that account.[8] California also enacted SB 73 in May 2026, a law directed at the ballot-seizure conduct at issue.[9]

Those facts help explain why the California Supreme Court case has drawn attention. They do not change the professional-responsibility classification. The AI-citation issue stands or falls on the filing record and the reported oral-argument admission, not on whether a reader agrees with the sheriff, the Attorney General, the Legislature, or the election-law challengers.

Current record status

As of Aug. 26, 2026, Cervantes v. Bianco, S295866, and Attorney General v. Bianco, S295901, should be carried as a live California Supreme Court ballot-seizure record with a confirmed AI-citation-fabrication risk component. The confirmed risk component is the chain from the filed challenge that the Common Cause quotation appears nowhere in the opinion to the press-reported oral-argument admission that AI was used and the citations were not verified before filing.

The record should not be described as an AI sanctions case unless the California Supreme Court or another authority later imposes a sanction, orders corrective action, refers counsel, or otherwise makes a finding. It should be flagged for update when the court issues its decision, when an official transcript or recording is available, or if any corrective filing, sanction motion, sanction order, or disciplinary referral appears.

References

1. Was sheriff’s seizure of 650,000 ballots legal? California Supreme Court weighs case — California Courts Newsroom
2. California Redistricting Referendum Ballot Seizure Challenge — Cervantes — Democracy Docket
3. Supreme Court Grants Review, Stays Investigation in Sheriff Ballot Seizure Case — Horvitz & Levy
4. California sheriff used non-existent quotes in legal defense of ballot seizures — Democracy Docket
5. California Supreme Court Puzzled by GOP Sheriff’s Bizarre Defense for Seizing Ballots — Democracy Docket
6. California high court questions Riverside County sheriff’s ballot investigation — Courthouse News
7. ABA issues first ethics guidance on a lawyer’s use of AI tools — American Bar Association, July 29, 2024
8. Baseless Allegations Drive Law Enforcement Seizures of Election Records — Brennan Center
9. California ballot seizure law — CalMatters