Rodrigues v. Craig: AI-Fabricated Citations in a Family Law Appeal Draw Appellate Court Rebuke | Advocate Prakhar

Case at a Glance

Full Citation	In re Marriage of Rodrigues & Craig, Nos. H051992, H052487, H052559, 2026 WL 1102361 (Cal. App. 6th Dist. Apr. 22, 2026)
Court	Court of Appeal, Sixth Appellate District, California
Date	April 22, 2026
AI Tool	AI tool (unspecified); nine fabricated case names identified in the opinion
Outcome	All trial court orders and judgment affirmed; fabricated citations catalogued and noted as potential grounds for frivolous-appeal dismissal; parties to bear own appellate costs

Background

Joanne Rodrigues and Nathan Craig were married for six years and eight months before Rodrigues petitioned for dissolution in February 2020. After a prolonged court trial spanning multiple dates between July 2022 and July 2024, the Santa Clara County Superior Court issued judgments dividing community property, denying Rodrigues’s separate property claim, setting spousal support at zero, and sanctioning Rodrigues for multiple breaches of fiduciary duty and for violating Automatic Temporary Restraining Orders (ATROs). The sanctions included $10,000 for breach of fiduciary duty, $25,000 for repeatedly refusing to participate in settlement conferences, and $7,500 for relocating the couple’s children to Newark without court authorization.

Rodrigues, a highly educated self-represented litigant — holding degrees from Georgetown University, UC Berkeley, and the London School of Economics — filed three separate appeals challenging the division of community property, the denial of her separate property claim, the denial of spousal support, and the sanctions. The three appeals were consolidated. The critical problem: the appellant’s brief submitted to the Court of Appeal contained citations to cases that do not exist, cases that California courts can no longer follow because they were disapproved by the Supreme Court, and cases that do not support the propositions for which they were cited.

The court’s editorial note at the start of the Westlaw report is itself unusual: “This decision contains discussion of citation references that are incorrect or do not actually exist. These invalid citations appeared in the original court opinion and have been preserved as written since they are part of the official record.” The court went out of its way to name the fabricated cases explicitly in the body of the opinion itself.

The AI Issue

The core AI-related legal question was whether a self-represented appellate litigant’s submission of briefs containing fabricated case citations — apparently generated by artificial intelligence — constitutes the kind of conduct that can independently justify dismissing an appeal as frivolous, separate from the merits of the underlying claims. The court also addressed the consequences for the overall credibility of the brief and the effect on appellate review when a substantial portion of the cited authority is non-existent.

What the Court Decided

• The court identified nine fabricated case citations in Rodrigues’s brief, naming each one: In re Marriage of Brewster, In re Marriage of Dorn, In re Marriage of Freeman, In re Marriage of Gowan, In re Marriage of Mele, In re Marriage of Olson, In re Marriage of Rivera, In re Marriage of Spector, and Moore v. California State Bd. of Equalization — none of which appear to exist.
• The court also found that Rodrigues cited two cases — Hall v. Harker and Catchpole v. Brannon — that the California Supreme Court had expressly disapproved, meaning they can no longer be cited as authority.
• The court held that extensive reliance on nonexistent legal authority “could alone justify dismissing an appeal as frivolous” under California Rules of Court, citing Noland v. Land of the Free, L.P. (2025) for the proposition that a brief peppered with fabricated citations unreasonably violates the Rules of Court.
• The court nonetheless addressed all of Rodrigues’s substantive arguments and affirmed every challenged order and the final judgment, finding no abuse of discretion in the community property division, sanctions, or denial of spousal support.
• Because the record was incomplete (Rodrigues omitted nearly all of respondent’s filings and exhibits from the appendix), the court applied the doctrine of implied findings and presumed the trial court made all necessary findings supported by substantial evidence.

“Extensive reliance on nonexistent legal authority could alone justify dismissing an appeal as ‘frivolous because it rests on negligible legal foundation’ and is peppered with fabricated legal citations… [and] unreasonably violates the Rules of Court because it does not support each point with citations to real (as opposed to fabricated) legal authority.”

— Court of Appeal, Sixth District, California, April 22, 2026

The India Angle

Indian Law Equivalent

In India, submitting non-existent case citations to any court — trial, High Court, or Supreme Court — would violate the advocate’s duty of candor enshrined in the Supreme Court’s inherent powers under Article 142 of the Constitution and the Bar Council of India Rules. The Code of Civil Procedure, 1908, Order VI Rule 16 permits striking out vexatious or misleading pleadings. In family courts specifically, the Family Courts Act, 1984 requires parties and their counsel to assist the court in reaching a just decision, not mislead it with invented authority. Courts have the power under the Contempt of Courts Act, 1971 to take cognizance of any filing that scandalizes or tends to scandalize the administration of justice.

Bar Council Rules

Under Rule 15 of Part VI, Chapter II of the Bar Council of India Rules, an advocate must not knowingly make a false statement of law or fact to the court. Rule 20 prohibits an advocate from misleading the court by referring to decisions or authorities not applicable. Submitting AI-generated fictitious case names in a family court brief — whether knowingly or through negligent failure to verify — would constitute professional misconduct under Section 35 of the Advocates Act, 1961, potentially resulting in suspension or removal from the rolls. The Bar Council of India has taken a conservative stance on technology use in practice, and Indian advocates should not assume that AI errors will be treated differently from deliberate fabrications.

Practical Advice for Indian Advocates

• In family law matters, where parties are often emotionally invested and self-represented litigants increasingly use online AI tools to draft petitions and written submissions, advocates reviewing or verifying those documents must treat every cited case as unverified until confirmed — the client’s AI cannot be trusted for citation accuracy.
• When using any AI tool to draft arguments for High Court or Supreme Court appeals, run every citation through SCC Online, Manupatra, or the official e-Courts database before filing; the nine fabricated family law cases named in this opinion show that AI tools frequently hallucinate plausible-sounding domestic relations case names that do not exist.
• If you discover that a citation in a draft brief is non-existent, delete it and its associated argument rather than substituting a real case that only partially supports the point — a poorly supported real argument is preferable to any argument resting on fabricated authority.

Quick Takeaways

• California appellate courts will catalogue AI-fabricated citations by name in published opinions, creating a permanent record of the false authority submitted — a significant professional and reputational risk for any advocate.
• The court confirmed that fabricated citations are independently sufficient to justify dismissal of an appeal as frivolous, regardless of whether the underlying claims have merit.
• Self-represented litigants who use AI to research and draft appellate briefs are held to the same citation accuracy standards as lawyers — education and intelligence offer no protection against AI hallucination liability if the output is not verified.

---

Deep Dive: AI Hallucinations in Family Law Appeals — A Growing Crisis

Family law litigation has historically been one of the areas of practice most reliant on self-represented litigants. Divorce, child custody, spousal support, and property division proceedings attract large numbers of parties who cannot afford counsel throughout the entire life of a case — and who therefore turn to whatever legal research resources they can access, including, increasingly, AI chatbots and AI legal research platforms. Rodrigues v. Craig is a landmark data point because it shows what happens at the appellate level when a highly educated, articulate, and persistent self-represented litigant places too much trust in AI-generated citations without verification.

Rodrigues is not a naive litigant. She holds graduate degrees from some of the world’s most respected institutions, has authored technical books in mathematics, and demonstrated considerable skill in representing herself across years of complex trial court proceedings. Yet her appellate brief contained nine case names that appear to have been entirely invented by an AI tool — names that follow the plausible California family law citation format (In re Marriage of [Surname]) perfectly, making them nearly indistinguishable from real cases at first glance. The court named all nine in the opinion, creating a permanent searchable record. It also noted two cases that were cited for propositions they cannot support because the California Supreme Court has disapproved them — a separate error that suggests the AI tool’s training data was not updated to reflect changes in case law status.

The court’s treatment of these errors is instructive for practitioners everywhere. Rather than using them as a procedural shortcut to avoid deciding the merits, the California court chose to address the fabricated citations directly and transparently — naming them, citing the authority for why they justify a frivolousness finding, and then proceeding to the merits anyway. The message is calibrated: courts will not allow AI hallucinations to help parties escape review, but they will make clear on the public record that the conduct is unacceptable. The result for Rodrigues was that her legitimate substantive claims — some of which appeared to have arguable merit — were evaluated and rejected, but they were evaluated against a backdrop of discredited authority that undermined the entire brief’s credibility.

For Indian family law practitioners, the lesson cuts in multiple directions. At one level, it is a warning about AI research in appellate practice. At another, it highlights the specific vulnerability of family law appeals, which typically turn on findings of fact reviewed deferentially — meaning an incomplete record combined with fabricated authority is a near-certain path to affirmance. Indian family courts increasingly deal with parties who have used online AI tools to generate petitions, maintenance applications, and custody submissions. Advocates who receive clients with AI-drafted documents must verify not just the law but also the procedural posture of every case cited. The risk is not merely professional — it is strategic. A brief undermined by hallucinated citations loses credibility on every point, including the points that might have been won.