What this record covers
AI in the Courtroom records matters in which an artificial intelligence system was used in litigation and a court, a rulemaker or a regulator then did something about it. It covers lawyers' filings, expert reports, the evidence itself, judges' own orders, what was shown or said in a courtroom, and the rules written in response. It does not cover lawsuits brought against the companies that build these systems. Those cases are about AI. The matters here are about how professionals and courts used it.
The distinction matters more than it sounds. A suit against a model developer turns on what the company did when it built the system. A matter here turns on what a professional did when using one, which is the only question an expert, a lawyer or a judge can do anything about in advance.
The record is organized into six topics: expert witnesses and ai; fabricated citations and facts in filings; AI-generated and altered evidence; judges, courts and ai; AI inside the courtroom; rules, orders and guidance. Every entry states what the professional or party did, what the tribunal did, where the matter stands and the date the entry was last checked. Each also names the clause of the Expert Record Standard that the conduct engaged. That last field is the one place the record goes beyond its sources. It describes conduct. It does not judge the person.
How this differs from the case record
The case record carries only the matters the book describes, and a matter enters it only after the order or opinion has been read in full. This section is broader and it moves faster. Many matters are reported days before the order is public, and some orders never are. So the record here is built on two kinds of source and every entry says which one it rests on.
An entry labeled Primary document read was checked against the order, opinion, rule or published guidance itself. An entry labeled Reported; document not located rests on reputable reporting because the underlying document had not been located on the day it was checked. The label appears on the entry, in every index row and in the topic lists. Of the 121 matters recorded so far, 116 were checked against the underlying document. When a document is later obtained the label changes and the entry's date advances. An entry that rests on reporting is not a lesser entry, but it is a different kind of claim, and a reader deciding whether to cite one needs to know which kind it is without having to work it out.
The six topics
Expert witnesses and AI
Expert declarations and reports prepared with an AI system, and how tribunals treated the evidence.
Fabricated citations and facts in filings
Filings that put a nonexistent case, quotation or fact before a court, and what the court did about it.
AI-generated and altered evidence
Deepfakes offered as evidence, AI-enhanced media, and claims that genuine recordings were fabricated.
Judges, courts and AI
Orders containing AI errors, judges using AI openly, and the oversight that followed.
AI inside the courtroom
Avatars, generated statements, AI in place of counsel, and AI tools used in proceedings.
Rules, orders and guidance
Rules of evidence, court rules, standing orders, ethics opinions and expert-facing guidance on AI.
Guides drawn from the record
Each guide answers one question that cuts across the six topics, from the matters in the record, and links every matter it relies on. The full list is on the guides page.
AI Disclosure and Certification Rules, Compared
Fifteen instruments in the record ask filers either to say that AI was used or to certify that its output was checked. They differ on who carries the duty, what it covers and whether experts are reached at all.
Deepfake Claims About Evidence: How Tribunals Have Ruled
Fabricated exhibits and the claim that real evidence is fabricated have both reached tribunals. The responses differ, and the record shows what each tribunal asked for.
Discovery of AI Prompts and Chat Logs: What the Record Shows
A small, growing set of orders addresses whether an expert's or a party's exchanges with a generative system must be produced. The outcomes differ, and nearly all are trial-level rulings.
Expert Reports That Cite Sources That Do Not Exist
Eighteen matters in the record show what tribunals did when expert material cited sources that could not be found or output nobody could verify. The outcomes differ, and the differences are instructive.
Proposed Rule 707: AI Evidence and the Daubert Standard
Many articles say AI evidence must now meet Daubert under a new Federal Rule of Evidence 707. The proposal is real, but it was withdrawn in the form published and a revised draft is still under study. This guide sets out the text, the record behind it and where it stands.
Every matter, most recent first
The full record in one table. Each row links to its entry, and each entry names its topic and the other matters filed beside it.
| Date | Matter | Who used AI | Outcome | Primary document read |
|---|---|---|---|---|
| 2026-11-12 | In re A.S., R.S.-1, and L.S. (W. Va.) | Appellate counsel for parents in an abuse and neglect appeal | Termination order vacated and remanded on the merits; brief referred to the Office of Disciplinary Counsel | ✓ |
| 2026-11-09 | Evans v. Collins (Federal Circuit) | Appellate attorney (veterans benefits) | Appeal dismissed for lack of jurisdiction; warning to the bar; no sanction | ✓ |
| 2026-11-09 | Gill v. Everyday Dose, Inc. | Civil attorneys (two pro hac vice counsel and Delaware counsel) | $3,000 Rule 11 sanction on each of two attorneys; order served on their licensing authorities; contempt facts certified as to one; no sanction on Delaware counsel | ✓ |
| 2026-11-04 | Gillespie v. Superior Court (People) | Prosecution expert (forensic psychologist) | Writ petition denied; expert opinion not excluded; trial court's credibility finding left undisturbed | ✓ |
| 2026-10-30 | Quanta Logistics v. Commissioner of Customs (Import-I) | Customs adjudicating authority | Customs order quashed by consent and remanded for a fresh order | ✓ |
| 2026-10-28 | Farhat v. Cutera, Inc. | Civil defense attorney | Show-cause order discharged; no sanction; counsel admonished; challenged and replacement authorities disregarded | ✓ |
| 2026-10-23 | South African Post Office v Commissioner Fakude | Judge | Arbitration stayed and interdicted pending review; no order as to costs | ✓ |
| 2026-10-21 | Fontainebleau Florida Hotel v. Botach | Party | AI-provider subpoenas and document request narrowed to 3 subjects; blanket work-product objection overruled; motion to compel consent denied | ✓ |
| 2026-10-21 | HHS and NTEU, 74 FLRA No. 87 | Party | Agency exceptions denied; caution on AI use; no sanction | ✓ |
| 2026-10-19 | Commonwealth v. Butler (Pa. Super. Ct.) | Defense attorney | Judgment of sentence affirmed; counsel reminded of ethical obligations; no sanction ordered | ✓ |
| 2026-10-15 | Deepak Raheja v. Omkara Assets Reconstruction | Civil attorney | Appeals dismissed on merits; proposed exemplary costs deleted; no order as to costs | ✓ |
| 2026-10-15 | Osteoarthritis Right Shoulder Review, 2026 CanLII 116891 | Medical opinion provider (nurse practitioner) | Medical opinion given no weight; entitlement granted on other evidence | ✓ |
| 2026-10-08 | Avergil v. Yaakov | Expert witness | Expert report struck; amendment motion dismissed; costs to each third party | ✓ |
| 2026-10-07 | Kid International v. City National Bank of Florida | Civil attorney | Summary judgment affirmed; appellate attorney's fees awarded as a sanction, jointly and severally | ✓ |
| 2026-09-30 | California SB 574 (Statutes of 2026, chapter 858) | Legislature | Enacted (chapter 858, Statutes of 2026) | ✓ |
| 2026-09-29 | Villanueva v. Las Vegas Metropolitan Police Department | Expert witness | ChatGPT case summaries ordered produced; prompts, outline and draft output protected | ✓ |
| 2026-09-25 | Aguilar v. The Crawford Group, Inc. | Civil attorney | Rule 11 sanctions; fees up to $10,000; pro hac vice revoked | ✓ |
| 2026-09-24 | Cervantes v. Bianco (Supreme Court of California) | Law firm | AI use disapproved; matter referred to the State Bar | ✓ |
| 2026-09-22 | TRAX International Corp. v. United States | Party | Award upheld; AI evaluations added to the record; government counsel ordered to show cause | ✓ |
| 2026-09-22 | Zhejiang Dingli Machinery Co. v. United States | Law firm | Order to show cause issued; no finding of wrongdoing; no ruling located | ✓ |
| 2026-09-18 | Tenth Circuit proposed Rule 46.5(B)(4) on generative AI | Court | Proposed; public comment open | ✓ |
| 2026-09-16 | Lisandrillo v. Palozzi | Civil attorney | Petition denied; order to show cause why sanctions, such as Florida Bar referral, should not be imposed | ✓ |
| 2026-09-14 | Henry v. Long Island University | Civil attorney | Two-year requirement to file the opinion in every EDNY appearance; no monetary sanction | ✓ |
| 2026-09-11 | Virgil v. Experian Information Solutions | Civil attorney | $10,000 personal Rule 11 sanction; separate referral to the Indiana Attorney Disciplinary Commission | ✓ |
| 2026-09-09 | Youssian v. Village of Kings Point | Civil attorney | $5,000 sanction; $10,000 in fees; petition and amended petition struck; proceeding dismissed | ✓ |
| 2026-09-03 | Assemblies of God Gospel Church v. Registrar of Societies | Civil attorney | Advocate suspended six months; public apology ordered; cases expunged | ✓ |
| 2026-09-03 | Douglas v. Deutsche Bank National Trust Co. | Civil attorney | Appellee's brief struck; referral to the Office of Disciplinary Counsel | ✓ |
| 2026-09-02 | Vijay Ghanshyam Gadiya v. Union of India | Court | Penalty order and High Court judgment set aside; remanded to a different officer | ✓ |
| 2026-08 | Watson Grinding explosion trial (3M expert) | Expert witness | Examined at trial on ChatGPT logs; no ruling on AI use | Reported |
| 2026-08-31 | AI clauses in protective orders (Jeffries; Dunn) | Court | AI provisions entered in protective orders | ✓ |
What an entry will not tell you
No entry characterizes a holding. An order that fines a lawyer $5,000 is described as fining a lawyer $5,000. It is not described as establishing a rule about AI, because a single order rarely does and because that argument belongs in the book, where a reader can see it and disagree with it. Nothing here is legal advice. Whether any matter bears on your own is a question for counsel.
The AI in the Courtroom record is not complete and does not try to be. One public tracker of decisions that address hallucinated content (material an AI system presented as real that does not exist) listed more than 2,000 by September 2026. A list that repeated all of them would bury the matters worth studying. What appears here was selected because it shows a distinct pattern, involves an expert or a court, or produced a written ruling other tribunals are likely to read.
Where the disagreements are
Two questions are unsettled enough that an entry on one side of either should not be treated as the law. The first is whether a party's own session records with a system are protected from discovery. Three trial courts have now answered differently, and no court of appeals in the United States has answered at all. The second is whether an expert's prompts form part of the method and are therefore discoverable, which one magistrate judge has held and which remains stayed on objection. Both questions are tracked here through the individual matters rather than summarized, because a summary would suggest a settled position that does not exist.
On the first question the positions are now stated plainly enough to name. Morgan v. V2X held that a self-represented party's use of a system falls within work product, because such a party is simultaneously the party and the advocate, while still ordering the platform named. Warner v. Gilbarco held that using a system is not itself a waiver, since waiver of work product requires disclosure to an adversary. Tym v. Cerno followed Morgan and required a privilege log. United States v. Heppner went the other way on a represented defendant's documents, and Shealy v. Seaside Investments went the other way again where the prompting was done by a party's partner with no attorney directing it. Two courts declined to follow Heppner on consecutive days in June 2026, under work-product rules worded differently from the federal one. The practical consequence is that the answer depends on the forum and on who did the typing.
A third question is forming behind those two. In Fry v. Fry a court declined to seal material a party had already uploaded to a consumer system, reasoning that the information had gone to a private company and was no longer confidential. That is not a discovery question at all. It is whether confidentiality survives the act of asking.
Reading a new report in this area
Most reports travel as a headline and a quotation. Before relying on one, establish 4 things. Whose document contained the problem: a lawyer's brief, an expert's declaration, a party's exhibit or the court's own order. What the tribunal actually did, in its own terms. Whether the matter is final. And whether the account comes from the order or from someone describing it. A dozen articles summarizing one order are one source.
Then read for conduct rather than for rule. What transfers from one matter to the next is the sequence of steps a professional took and the point at which that sequence could no longer be defended. The disposition turns on the posture and on the record before that tribunal, and it will rarely match your situation. The eight obligations that sequence has to satisfy are set out in the Standard.
How the record is kept
Entries are added to AI in the Courtroom as matters are reported and verified, and they are revised when a matter moves: an appeal is decided, a sanction is imposed, an order is withdrawn or a rule takes effect. Each entry carries the date it was last checked, which is a different fact from the date it was added. A matter that is vacated or withdrawn stays here with its new posture, because in this area the reversal is often as instructive as the original order. That differs from the case record, where a vacated matter comes off the page and the removal is noted in the errata.
How the research is done, including the use of AI systems in it, is set out on the methodology page. If you know of a matter that belongs here, or an entry that is out of date, write to [email protected].
Questions about this record
Is using AI in a court filing a violation in itself?
No entry in this record shows a sanction for using AI as such. The sanctions recorded here follow from what was filed without being checked, such as a nonexistent case, an invented quotation or a misstated record. Some judges and courts do require disclosure or certification of AI use, and those requirements are collected under rules, orders and guidance.
Why are expert witnesses a separate topic?
Because an expert's position differs from counsel's. The expert's work is evidence, and the consequences in the expert matters recorded here have mostly taken the form of exclusion of that evidence rather than sanctions on the expert. It is also the subject of the book.
What does the label Reported; document not located mean?
It means the order, opinion or other primary document had not been located and read when the entry was last checked, so the entry rests on the reporting it lists. The label is removed when the document is found and read.
How current is AI in the Courtroom?
Each entry shows the date it was last checked, and pending matters say they are pending. The date on an entry tells you when it was last true, which is the only date that matters for relying on it.
Can I cite an entry?
Cite the underlying document. An entry exists to help you find that document and to tell you whether it is still good. It is not a substitute for reading it.