Amir Mostafavi: 21 Fake ChatGPT Citations and $10,000 in Sanctions | Advocate Prakhar

Case at a Glance

Background

Amir Mostafavi, a California-licensed attorney, became one of the most frequently cited examples of repeat AI hallucination misconduct in US legal history. Across multiple civil cases filed in California courts, Mostafavi submitted briefs containing case citations generated by ChatGPT — citations to cases that did not exist, had different holdings than stated, or were attributed to the wrong courts.

What distinguishes the Mostafavi matter from other AI hallucination cases is the volume — 21 fictitious citations identified across his filings — and the fact that misconduct continued after courts had already flagged the problem. This pattern transformed the characterisation of the conduct from negligent to potentially wilful.

The AI Issue

The core legal question across the Mostafavi proceedings was whether an attorney who repeatedly submits AI-generated fictitious citations — particularly after receiving judicial warnings — can be said to have violated professional conduct rules with the requisite mental state for enhanced sanctions and potential bar referral. The cases also raised questions about whether pattern misconduct across multiple courts should be treated cumulatively for sanction purposes.

What the Court Decided

• Courts imposed $10,000 in aggregate sanctions across the Mostafavi proceedings for submitting fabricated ChatGPT citations [Rule 11 / California Rules of Court, Rule 2.30 — sanctions for frivolous filings].
• Continuation of AI citation errors after judicial warnings indicated a pattern of conduct beyond mere inadvertence.
• Courts ordered Mostafavi to file declarations certifying personal verification of all future legal citations in any California court proceeding.
• The pattern of conduct across 21 fictitious citations in multiple cases was treated as an aggravating factor warranting higher sanctions than a single-incident case.
• The matter was noted for potential referral to the California State Bar for professional conduct review.

“The repeated submission of non-existent citations, even after the court’s prior admonishments, cannot be characterised as an isolated error. Counsel has demonstrated a pattern of conduct that this court cannot overlook.”

— California Superior Court (paraphrase), 2025

The India Angle

The Mostafavi case raises a question Indian advocates must confront: what happens when AI citation errors recur, particularly as AI tools become standard in legal research workflows?

Indian Law Equivalent

In India, the Contempt of Courts Act, 1971 distinguishes between civil contempt (wilful disobedience of a court order) and criminal contempt (scandalising the court). An advocate who continues to cite non-existent cases after a court’s explicit warning could face criminal contempt proceedings — the continuing nature of the conduct being the key aggravating factor, exactly as in the Mostafavi matter. Additionally, under the Advocates Act, 1961, the Bar Council of India and State Bar Councils have disciplinary jurisdiction over misconduct by advocates. Persistent fabricated citations could constitute professional misconduct warranting suspension or disbarment under Section 35 of the Advocates Act.

Bar Council Rules

Rule 22 of the BCI Standards requires advocates to act with dignity and self-respect in the conduct of their cases. Persistent citation of non-existent cases — especially after judicial warning — is plainly inconsistent with this standard. State Bar Councils have the authority to initiate suo motu disciplinary proceedings on the basis of a court’s order or remarks.

Practical Advice for Indian Advocates

• Treat any judicial observation about a citation error as a serious warning requiring immediate remedial action — never repeat the same category of error in the same or any other proceeding.
• If AI tools are used for research, maintain a written log of every citation used and the source (official database) that verified its existence — this log can be produced to court if your methodology is questioned.
• Where a court has previously raised concerns about AI use in your filings, consider proactively disclosing your verification process in subsequent submissions to demonstrate good faith compliance.

Quick Takeaways

• 21 hallucinated citations across multiple cases — volume and repetition are aggravating factors.
• Continuing AI errors after judicial warnings transforms negligence into potential bad faith.
• Pattern misconduct risks both escalating sanctions and State Bar referral.

Deep Dive: Repeat Misconduct and the Escalating Sanctions Framework

Why 21 Citations Matters

Most AI hallucination cases that resulted in sanctions involved a discrete set of fictitious citations — typically between 6 and 12 — in a single proceeding. The Mostafavi matter stands out because the count of 21 fictitious citations spans multiple proceedings and extends across a period after courts had already put the attorney on notice. This is the feature that elevates the Mostafavi case from a cautionary tale about negligent AI use to a case study in escalating professional accountability.

The legal significance of the number is not merely rhetorical. Courts applying Rule 11 (federal) or its state equivalents conduct a proportionality analysis when calibrating sanctions — the more egregious and extensive the misconduct, the higher the appropriate sanction. Volume is an objective indicator of severity.

The Concept of Constructive Knowledge After Warning

A key doctrinal development in the Mostafavi proceedings is the court’s treatment of a prior judicial warning as imputing constructive knowledge about the risk of AI hallucination. Once an attorney has been warned — formally or informally — that AI tools generate fictitious citations, any subsequent reliance on unverified AI citations cannot be characterised as innocent inadvertence. The attorney is on notice of the specific risk and assumes the legal consequences of proceeding without adequate safeguards.

This principle has direct implications for any attorney who has read about AI hallucination sanctions (including the extensively reported Mata v. Avianca case from 2023) — courts may increasingly argue that awareness of the general phenomenon of AI hallucination, even absent a specific personal warning, is sufficient to support a finding of recklessness.

The California Bar Context

The California State Bar has been one of the most active bar authorities in issuing guidance on AI use in legal practice. Following a series of AI hallucination sanctions in California courts, the Bar released formal guidance in 2024 addressing competence obligations when using AI tools — drawing on Rule 1.1 of the California Rules of Professional Conduct (Competence) and its obligation to stay current with “changes in the law and its practice, including the benefits and risks associated with relevant technology.”

The Mostafavi pattern is precisely the kind of conduct that the California Bar’s guidance was designed to prevent. Whether bar discipline proceedings are ultimately initiated against Mostafavi, and what outcome those proceedings produce, will be an important data point for the broader question of how bar regulators treat AI-related professional misconduct.

Global Context: When Does Misconduct Become Disqualifying?

The escalating severity of AI hallucination consequences tracks a pattern seen in other contexts of technology-related professional misconduct. Courts and bar regulators are working through a spectrum: from civil sanctions (monetary) to referral to bar authorities, to suspension, to disbarment. The Mostafavi case sits toward the more serious end of the spectrum but has not yet reached the disqualifying extreme seen in the Alabama Trust case, where an attorney was barred in connection with AI citation misconduct that caused an appeal to fail.

For Indian practitioners watching this trajectory, the lesson is clear: the window for treating AI hallucination as a minor procedural irregularity is closing. Courts and bar bodies globally are establishing that it is a professional conduct issue with serious career consequences.