The Legal AI Judicial Clerk Influence Report 2026: How Law Clerks at Federal District and Circuit Courts Are Actually Encountering, Flagging, and Informally Shaping Judicial Responses to AI-Assisted Filings

Research briefing — a review of public sources. The Legal Stack did not conduct a survey, benchmark or interviews for this piece.

What the public record can and cannot tell us

The question in the title concerns how federal law clerks encounter, flag and informally shape judges' responses to AI-assisted filings. The public record cannot answer it directly. No survey of federal clerks on this subject has been published. Chambers deliberations are confidential, and courts do not record when a clerk raises a concern about a brief. Any claim about how often clerks "suspect" AI text, or how they tell their judges, would be speculation presented as data. This briefing makes no such claims.

What is documented is the following:

the volume of AI-related citation failures that reach judicial decisions;
the patchwork of standing orders and rules governing AI disclosure;
the federal judiciary's own interim guidance on AI use inside chambers;
two 2025 episodes in which AI use by chambers staff produced flawed opinions.

Taken together, these show what kind of review a filing faces in chambers, which is what litigators actually need to know.

The visible record: hallucination cases keep growing

The formal record of AI-related filing failures is no longer small. Damien Charlotin's public AI Hallucination Cases database had logged about 700 court decisions by the end of 2025, and about 800 US cases by April 2026, as reported by Bloomberg Law and heise.de. The database covers courts at all levels, not only federal ones.

The founding federal example is Mata v. Avianca. On 22 June 2023, Judge P. Kevin Castel of the Southern District of New York sanctioned two lawyers $5,000 under Rule 11 for submitting fictitious ChatGPT-generated citations. He also required them to send the opinion to their client and to each judge falsely named as the author of a fabricated decision. At appellate level, in Park v. Kim, 91 F.4th 610 (2d Cir. 2024), the Second Circuit referred a lawyer to its grievance panel over an AI-fabricated citation.

These cases are relevant to clerks because someone had to find the fabricated authorities, whether opposing counsel, a clerk or the judge. A citation that does not exist is the easiest AI failure to detect and the most likely to be caught at the cite-checking stage, which in many chambers is clerk work.

The rule map is patchy, and that is deliberate

Practitioners often overstate how far courts have gone in formal AI rules. Bloomberg Law's tracker of judicial standing orders on AI counted 39 orders from federal judges between May 2023 and April 2025:

35 came from district judges;
3 came from bankruptcy judges;
1 came from the Court of International Trade.

Orders were spread across about 20 district courts, with the Northern District of California issuing the most. That is a small fraction of the federal district courts. Bloomberg Law also observed signs that judges were shifting from issuing standing orders towards sanctioning lawyers.

At circuit level, the Fifth Circuit considered a rule that would have required lawyers to certify AI use and human verification of citations. On 12 June 2024 it declined to adopt it and relied instead on existing duties under Rule 11 and the duty of candor. The Judicial Conference has not required certification of AI-assisted filings. Its AI rulemaking work has centred on proposed Federal Rule of Evidence 707, which concerns machine-generated evidence rather than briefs. According to Proskauer:

the rule was published for comment from 15 August 2025 to 16 February 2026;
in June 2026 the Standing Committee decided not to recommend it, sending it back for revision and further study.

State courts are moving faster in places. On 28 May 2026, the Florida Supreme Court amended Rule of General Practice and Judicial Administration 2.515 to create a statewide citation and AI-verification certification requirement.

Analysis. Where no disclosure rule exists, the operative standard in federal court is the general one: Rule 11 and candor. A filing is judged by whether its citations and record references hold up, not by whether AI was used. That is a standard clerks are well placed to apply.

Inside chambers: the judiciary's own AI guidance

The most concrete public evidence about AI and chambers staff comes from the courts themselves. On 31 July 2025, the Administrative Office of the US Courts distributed interim AI guidance to federal courts, developed by a task force formed earlier that year. Director Judge Robert J. Conrad described it in a letter to Senator Chuck Grassley released on 23 October 2025. FedScoop reported that the judiciary did not publish the full document. According to that reporting, the guidance:

applies to judges, staff and other judiciary personnel;
warns against delegating "core judicial functions to AI, including decision-making or case adjudication";
urges "extreme caution" on novel legal questions;
requires users to verify AI outputs independently, since they "are accountable for all work performed with the assistance of AI";
leaves decisions about disclosure to the courts and lets each court define which tasks locally approved tools may be used for.

The guidance followed two episodes that the ABA Journal reported on 24 October 2025. Responding to Grassley's questions, Judge Henry T. Wingate (S.D. Miss.) said a law clerk had used Perplexity. His 20 July temporary restraining order had cited allegations, parties and declarations that did not exist. Judge Julien Xavier Neals (D.N.J.) said a law school intern had used ChatGPT for research without authorization, contrary to chambers policy. His 30 June opinion misstated case outcomes and attributed fabricated quotations. Both opinions were withdrawn after litigants pointed out the errors. Both judges said the drafts should not have been docketed before completing normal review.

Analysis. These episodes are the only publicly documented cases we found that link AI use to specific chambers staff. They point the opposite way from the original framing of this article. The documented risk was not clerks policing lawyers' AI use, but clerks and interns using AI themselves. The judiciary's response was to treat chambers staff as AI users who need guardrails, not as gatekeepers.

What litigators can reasonably infer

Clerk practices vary by chambers and are not public. The following inferences are labelled as analysis.

Assume every citation will be checked. Cite-checking is routine chambers work. Since 2023, a non-existent case has been one of the most visible ways a brief can fail. The safest course is a filing in which every authority and pin cite has been verified by a person.
Tie argument to the record. In Wingate, the errors involved non-existent parties and declarations, not just cases. Claims about the record are as checkable as claims about case law, and the courts' interim guidance shows they expect verification of both.
Know the local rule, then go beyond it. Check the Bloomberg Law tracker, or the individual judge's page, for a standing order before filing. Where there is none, the absence of an AI rule does not change Rule 11. The Fifth Circuit relied on exactly that reasoning in declining its proposed rule.
Do not try to look non-AI. We know of no published evidence that clerks reliably detect AI-drafted prose by style, and no court treats stylistic features as misconduct. The documented sanctions concern false content. Effort spent on accuracy is worth more than effort spent on making prose sound human.

The real data gaps

Several questions in this area cannot be answered from public sources:

how often chambers staff raise concerns about filings that never become sanctions;
whether those concerns affect outcomes;
how individual courts have put the AO's interim guidance into practice.

Answering them would require the judiciary to publish its guidance and local implementations, or researchers to study chambers practice with courts' cooperation. Until then, commentary that describes clerks' private views as fact should be treated with caution, and that includes earlier versions of this briefing.

Sources

ABA Journal, "2 federal judges reveal AI use by staff members led to error-riddled opinions" (24 Oct 2025): https://www.abajournal.com/news/article/2-federal-judges-reveal-ai-use-by-staff-members-led-to-error-riddled-opinions
FedScoop, "Interim AI guidance for US courts aims for experimentation, guardrails" (24 Oct 2025): https://fedscoop.com/interim-ai-guidance-us-courts-aims-experimentation-guardrails/
Administrative Office of the US Courts, letter to Sen. Grassley re judiciary use of AI (Oct 2025): https://www.grassley.senate.gov/imo/media/doc/ao_to_grassley_re_judiciary_use_of_ai.pdf
Bloomberg Law, "Which Federal Courts Have AI Judicial Standing Orders" (16 Apr 2025): https://news.bloomberglaw.com/safety/which-federal-courts-have-ai-judicial-standing-orders
Proskauer, "Machine-Generated Evidence Challenges the Federal Rules: Inside the Contested Proposed Federal Rule of Evidence 707" (29 Jul 2026): https://www.proskauer.com/blog/machine-generated-evidence-challenges-the-federal-rules-inside-the-contested-proposed-federal-rule-of-evidence-707
Wikipedia, "Mata v. Avianca, Inc." (accessed Oct 2026): https://en.wikipedia.org/wiki/Mata_v._Avianca,_Inc.
Law360, "2nd Circ. Rebukes Atty For Fake Citation In Latest AI Blunder" (30 Jan 2024): https://www.law360.com/legalethics/articles/1791788
King & Spalding, "U.S. Fifth Circuit Decides Against Its Proposed Rule Amendment on AI Use in Legal Filings" (14 Jun 2024): https://www.kslaw.com/insights/articles/us-fifth-circuit-decides-against-its-proposed-rule-amendment-on-ai-use-in-legal-filings
Damien Charlotin, "AI Hallucination Cases" database (accessed Oct 2026): https://www.damiencharlotin.com/hallucinations/
Bloomberg Law, "AI-Faked Cases Become Core Issue Irritating Overworked Judges" (29 Dec 2025): https://news.bloomberglaw.com/us-law-week/ai-faked-cases-become-core-issue-irritating-overworked-judges
heise online, "AI hallucinations in court" (2026): https://heise.de/-11257675
Jones Day, "Supreme Court of Florida Creates Statewide Standard for Artificial Intelligence Certification in Court Filings" (Jun 2026): https://www.jonesday.com/en/insights/2026/06/supreme-court-of-florida-creates-statewide-standard-for-artificial-intelligence-certification-in-court-filings
Judicial Conference of the United States, absence of a rule requiring certification of AI-assisted filings (no public link located)