Glossary

The record on this site draws on courts in several countries, on rules of evidence and procedure, and on vocabulary from the technology itself. This glossary defines those terms in general form. Where a term comes from a particular legal system, the definition says so, and where a term is the Standard's own, it is defined as the Standard defines it.

Definitions describe how terms are used; they are not legal advice and do not state the law of any jurisdiction. How a rule applies in a given matter is a question for counsel. The methodology page explains how the record itself is researched.

A · B · C · D · E · F · G · H · I · L · M · P · Q · R · S · T · W

A

Attorney-client privilege

In US law, the protection that keeps confidential communications between a client and a lawyer, made to seek or give legal advice, from compelled disclosure; other common-law systems recognize a comparable legal professional privilege. Whether it reaches a client's own exchanges with an AI system is unsettled. In United States v. Heppner, a federal trial court rejected the claim because the system is not a lawyer and the provider's published policy undercut any expectation of confidentiality.

See: Discoverability of AI prompts and chat logs

Authentication

In US federal courts, Federal Rule of Evidence 901 requires the party offering an item to produce evidence sufficient to support a finding that the item is what that party claims it is; state evidence codes contain parallel rules. Claims that a recording or image is a deepfake are usually argued under this rule. In State v. Amyda (deepfake claim without evidence), the appellate courts held a video properly authenticated despite a deepfake claim made without evidentiary support.

See: Deepfake claims about evidence AI as evidence

B

Bar referral

A court's transmission of a lawyer's conduct to the body that licenses and disciplines lawyers in that jurisdiction. The referral is not itself discipline: the licensing body decides whether any follows, as it later did after the referral in Noland v. Land of the Free, L.P..

See: Fabricated authority

Best-evidence rule

In US evidence law, the rule (Federal Rules of Evidence 1001 through 1008 and state equivalents) that proving the content of a writing, recording or photograph generally requires the original. A duplicate is usually admissible to the same extent unless a genuine question is raised about the original's authenticity or admitting the duplicate would be unfair. In State v. Amyda (deepfake claim without evidence), the court found that the defense had offered conjecture rather than a genuine question.

See: Deepfake claims about evidence

C

Certification requirement

A rule or order requiring a person who files a document to state whether generative AI was used in preparing it and, often, to confirm that a human checked every citation. The record includes a single judge's certificate in Judge Starr's AI certificate and the Fifth Circuit rule and a two-year requirement imposed as part of a sanction in Lnu v. Blanche (Ninth Circuit suspension).

See: AI disclosure and certification rules Rules, orders and guidance

Chat log (AI history)

The stored record of the prompts a user entered into an AI system and the responses it returned, usually retained by the provider and exportable by the account holder. Courts have reached different results on whether an expert's chat log must be produced. In Villanueva v. Las Vegas Metropolitan Police Department, the court ordered the system's case summaries produced while protecting the expert's prompts and draft output.

See: Discoverability of AI prompts and chat logs Clause 7: Contemporaneous record

Contemporaneous log

In the Standard, the record an expert keeps as the work proceeds of each system used, the purpose of each use and what was done to verify the output, made at the time and capable of being produced. It is the obligation in Clause 7, and it is what makes the other clauses provable after the fact.

See: Clause 7: Contemporaneous record AI-use log template

Court-appointed expert

An expert chosen and instructed by the tribunal rather than retained by a party, serving as the court's neutral. It is the ordinary model in many civil-law systems, such as Germany's, and it also appears elsewhere, including in Israeli courts and, under Federal Rule of Evidence 706, in US federal courts. In the Landgericht Darmstadt expert fee decision, the court set such an expert's fee at zero after finding the report substantially AI-produced.

See: Expert witnesses Plonit v. Clal Insurance

D

Deepfake

Audio, video or an image generated or altered with AI so that it appears to record something that did not happen. In Mendones v. Cushman & Wakefield, a trial court found two video exhibits AI-generated and dismissed the case.

See: Deepfake claims about evidence AI as evidence

Draft Rule 901(c)

A draft amendment to the US Federal Rules of Evidence on alleged deepfakes, under study by the federal Advisory Committee on Evidence Rules. As described in the committee's materials, it would require the opponent to offer evidence sufficient to support a finding of fabrication, after which the proponent would have to show the item is more likely than not authentic. It had not been approved for adoption as of October 2026; the entry on Proposed Federal Rule of Evidence 707 tracks its status.

See: Deepfake claims about evidence Rules, orders and guidance

Draft-report protection

In US federal courts, Federal Rule of Civil Procedure 26(b)(4)(B), which shields drafts of a testifying expert's report or disclosure from discovery regardless of the form in which the draft is recorded; Rule 26(b)(4)(C) separately protects most communications between a retained expert and counsel. In Villanueva v. Las Vegas Metropolitan Police Department, the court treated AI output the expert used to draft his report as protected draft material.

See: Discoverability of AI prompts and chat logs

E

Expert Record Standard (the Standard)

The eight-clause standard for the use of artificial intelligence in expert engagements, issued as version 1.0 from the book The Expert Record by Bill Hartzer and licensed for verbatim adoption with attribution. Its clauses run from sequence to disclosure, and its governing test asks whether a qualified peer could reproduce the result without any AI system.

See: The Standard Pre-use checklist

F

Facts or data considered

In US federal courts, the category in Federal Rule of Civil Procedure 26(a)(2)(B)(ii) that a retained expert's written report must disclose: the facts or data the witness considered in forming the opinions, which is broader than what the report cites. Whether AI output falls within it is being litigated. In Villanueva v. Las Vegas Metropolitan Police Department, the court held that ChatGPT's case summaries were facts or data the expert considered.

See: Discoverability of AI prompts and chat logs Conservation Law Foundation v. Shell

G

Generative AI

Software that produces new content, such as text, images, audio or code, from patterns learned during training, in response to a request. The Standard's definition of a system covers it whether it is used directly or operates inside other software.

See: The Standard AI in the Courtroom

Governing test

The Standard's test for an engagement: whether a competent examiner in the same field, working only from the primary sources, with no access to any artificial intelligence system, could reproduce the expert's result. It tests the result rather than the process, so it permits use of these tools while ruling out dependence on them.

See: The Standard Clause 5: Reproducible analysis

H

Hallucination

Output from an AI system that is fluent and confident but false, most visibly a citation to a case, article or guideline that does not exist. The best-known example in the record is Mata v. Avianca, Inc., where a brief cited ChatGPT-produced cases that did not exist.

See: Fabricated authority Expert reports citing sources that do not exist

I

In camera review

A judge's private examination of disputed material, without the opposing party seeing it, to decide whether it is protected or must be produced; the term is used in US practice and in other common-law systems. In Villanueva v. Las Vegas Metropolitan Police Department, the district judge conducted such a review before dividing an expert's ChatGPT log into produced and protected parts.

See: Discoverability of AI prompts and chat logs

Independent authority and checkability

In the Standard, independent authority (Clause 3) means nothing enters a report on a system's word: each assertion is confirmed against a source that exists apart from the system, and the expert can name it. Checkability before adoption (Clause 4) is the step just before that, identifying the specific thing an output would be checked against and declining the output when nothing can be named.

See: Clause 3: Independent authority Clause 4: Checkability before adoption Verification checklist

L

Large language model

A type of generative AI trained on very large bodies of text to predict and produce language; ChatGPT, Claude and Gemini are built on such models. Their answers can vary between identical requests, a point recorded in the concurrences in Snell v. United Specialty Insurance (Judge Newsom).

See: Judges and courts Clause 5: Reproducible analysis

Liar's dividend

The possibility that public awareness of deepfakes makes it easier to cast doubt on genuine evidence. The term appears in State v. Amyda (deepfake claim without evidence), where the court used it to describe a defense argument resting on popular culture rather than evidence; Huang v. Tesla is an earlier civil instance of a party pointing to deepfakes to resist authentic recordings.

See: Deepfake claims about evidence

M

Motion to compel

In US civil procedure (Federal Rule of Civil Procedure 37(a) and state equivalents), a request asking the court to order a party or witness to provide discovery that has been withheld. Several AI disputes in the record began this way, including Conservation Law Foundation v. Shell (expert AI prompts), where a magistrate judge ordered an expert's prompts disclosed in an order later stayed pending objection.

See: Discoverability of AI prompts and chat logs

P

Pro hac vice

Permission for a lawyer not admitted in a court's jurisdiction to appear in a single case, usually alongside local counsel; the court that grants it may revoke it. In Aguilar v. The Crawford Group, Inc., a federal court revoked such an admission after filings with fictitious and misattributed citations.

See: Fabricated authority

Prompt

The text of a request put to an AI system, which the provider typically stores verbatim, in sequence and with a timestamp. Whether an expert's prompts are discoverable method or protected work is contested; compare Conservation Law Foundation v. Shell (expert AI prompts) with Villanueva v. Las Vegas Metropolitan Police Department.

See: Discoverability of AI prompts and chat logs AI-use log template

Proposed Rule 707

A proposed US Federal Rule of Evidence that would subject machine-generated evidence offered without an expert witness to the reliability requirements of Rule 702. Published for comment in August 2025, it was withdrawn as published and a modified draft was set for further study. No vote to publish a new version had occurred as of October 21, 2026; see Proposed Federal Rule of Evidence 707.

See: Rules, orders and guidance

Protective order

A court order limiting who may see material produced in discovery and what may be done with it, often in tiers such as confidential and attorneys'-eyes-only; in US federal courts it issues under Federal Rule of Civil Procedure 26(c). Some orders now address AI tools by name, as in AI clauses in protective orders (Jeffries; Dunn).

See: Clause 6: Custody and confidentiality Morgan v. V2X

Q

Quasi-judicial adjudicator

An official outside the court system, often within a government agency, who decides disputes with legal effect under procedures resembling a court's, such as a tax or customs adjudicating officer. In Vijay Ghanshyam Gadiya v. Union of India, the Supreme Court of India set aside a customs officer's penalty order that relied on authorities it found nonexistent or misapplied.

See: Judges and courts

R

Reproducible analysis

In the Standard, the obligation in Clause 5 that any analysis producing a number, count, sequence or selection be performed by a method that returns identical results when repeated, with the method itself preserved. In practice this usually means a script kept with its input manifest, raw output, environment notes and run date.

See: Clause 5: Reproducible analysis Analysis scripts

Rule 11

In US federal courts, Federal Rule of Civil Procedure 11, under which signing a filing certifies that its legal contentions are warranted and its factual contentions have evidentiary support after a reasonable inquiry; violations can draw sanctions. Many US sanctions for fabricated citations in the record rest on it or a state counterpart, as in Obi v. Cook County.

See: Fabricated authority

Rule 702

In US federal courts, Federal Rule of Evidence 702, which admits expert testimony only if the proponent shows it is more likely than not that the expertise will help the trier of fact, that the testimony rests on sufficient facts or data and reliable methods, and that the methods were reliably applied. Many states follow similar rules, while some apply the Frye general-acceptance test instead. The court in In re Celsius Network LLC (Faraj report) excluded an AI-generated expert report under Rule 702.

See: The Standard Expert witnesses

S

Show-cause order

An order directing a party or lawyer to explain why a sanction or other measure should not be imposed; it opens an inquiry and is not itself a finding of wrongdoing. Zhejiang Dingli Machinery Co. v. United States is an example in which no ruling on the order had been located.

See: Fabricated authority

Standing order

An order issued by a judge or court that applies to every case before it rather than to one matter, now often used to set requirements for AI-assisted filings. In Cherokee County Superior Court AI standing order, the judges of a state trial court required anyone who uses AI on a filing to name the tool and certify that the document was reviewed and every citation verified.

See: AI disclosure and certification rules

T

Temporary judge

In some US state court systems, a lawyer or court commissioner who hears a case as judge by the parties' stipulation. In In re Domestic Partnership of Torres Campos and Munoz, the appellate court held that a temporary judge's reliance on a nonexistent case was an abuse of discretion, though it found the claim forfeited.

See: Judges and courts

Terminating sanction

A sanction that ends a case, by dismissing the claims or entering default, generally reserved in US practice for serious misconduct where lesser measures would not suffice. In Mendones v. Cushman & Wakefield, a trial court imposed one after finding video exhibits AI-generated.

See: AI as evidence

W

Waiver

The loss of a privilege or protection, typically through disclosure to someone outside the protected relationship; in US law, voluntary disclosure to a third person generally waives attorney-client privilege, while work product is usually waived only by disclosure to an adversary or in a way likely to reach one. In Warner v. Gilbarco (ChatGPT work product), a federal court treated ChatGPT as a tool rather than a person and found no waiver of work product.

See: Discoverability of AI prompts and chat logs Villanueva v. LVMPD

Work product

In US federal courts, the protection under Federal Rule of Civil Procedure 26(b)(3) for documents and tangible things prepared in anticipation of litigation by or for a party or its representative, with heightened protection for mental impressions. Courts have divided on AI sessions: Morgan v. V2X (AI work product) recognized some protection for a self-represented party's AI use, while United States v. Heppner rejected it for documents a defendant generated without counsel's direction.

See: Discoverability of AI prompts and chat logs Tate Group v. Legacy