The Boundaries Of AI In Judicial Proceedings: From Parties To Lawyers

Abstract

With the increasing integration of AI into the legal field, the boundaries governing the use of AI tools by both parties and legal counsel have become increasingly prominent. From the perspective of parties, this article takes United States of America v. Bradley Heppner as a point of departure and examines the position that where a party, without the direction of counsel, uses consumer-grade AI platforms to process confidential information, communications between the party and the AI platform are not protected by the attorney-client privilege or the work product doctrine. From the perspective of lawyers, professional guidelines issued in jurisdictions such as Shanghai and Singapore, as well as existing cases, emphasize that lawyers bear an independent duty to review AI-generated content. The submission of false or inaccurate materials may violate duties of candor in litigation and may result in sanctions, including monetary penalties. Against this backdrop, parties can more reliably ensure the protection of case-related information only by obtaining legal advice through qualified counsel.

Main Text

In common law litigation, including in the United States and the United Kingdom, parties in civil proceedings are generally required to disclose materials relevant to the case through the discovery process¹. In criminal proceedings, prosecuting authorities may also require defendants to disclose documents or information². However, such disclosure obligations are not absolute. In certain circumstances, the law affords protection to particular communications or materials through privilege doctrines. In U.S. law, two principal doctrines serve to protect communications and materials between attorneys and their clients: the attorney-client privilege and the work product doctrine. The attorney-client privilege protects confidential communications between a client and counsel made for the purpose of obtaining or providing legal advice, while the work product doctrine protects materials prepared in anticipation of litigation³. These doctrines are rooted in the common law and have been developed through judicial decisions based on “reason and experience”, while also being subject to regulation under state law⁴. Within this framework, communications and materials that fall within the scope of these protections are generally exempt from disclosure in litigation. For example, the New York Civil Practice Law and Rules expressly provides that privileged matter and attorney’s work product are not subject to disclosure⁵, and similar provisions exist under Texas law⁶. In the United Kingdom, legal professional privilege likewise affords comparable protection to legal advice communications between lawyers and their clients, as well as to documents prepared for litigation⁷. It is precisely because of the existence of these privilege doctrines that parties are able to disclose case-related facts to their counsel without reservation during litigation.

I.The Boundaries of Parties’ Use of AI Tools in Judicial Proceedings

However, with the increasing use of generative AI platforms, parties confronted with legal issues often turn to AI tools for legal research or case preparation on their own initiative. This gives rise to a critical question: when a party discusses legal issues with an AI platform or inputs case-related information onto such a platform, can those communications still be protected by privilege?

On February 17, 2026, the United States District Court for the Southern District of New York, in United States of America v. Bradley Heppner⁸, examined the boundaries of a party’s use of AI tools in judicial proceedings. At a time when AI technologies are increasingly permeating all sectors, this case provides important guidance for clarifying the appropriate limits of AI involvement in judicial processes. The court held that written communications between a party and a generative AI platform are not protected by either the attorney-client privilege or the work product doctrine.

In November 2025, Bradley Heppner was arrested on charges including securities fraud and wire fraud. The Federal Bureau of Investigation (FBI) executed a search warrant at his residence and seized numerous documents and electronic devices. Among the seized materials were 31 reports generated by Heppner using Anthropic’s AI tool, Claude, which were created after he had received a subpoena but prior to his arrest, and without any direction from counsel. These reports contained detailed analyses of potential defense strategies and legal arguments.

The prosecuting authority sought to introduce these materials as evidence. Counsel for Heppner, however, argued that the documents were prepared for the purpose of facilitating communication with counsel and were subsequently provided to his attorney, and therefore should be protected by the attorney-client privilege and the work product doctrine. Ultimately, the court rejected the arguments advanced by counsel for Bradley Heppner and held that the AI-generated reports, which contain factual information, are admissible as evidence. The court’s reasoning is set out below.

To begin with, communications between a party and an AI platform are not protected by the attorney-client privilege. First, an AI platform is not a licensed attorney and therefore does not fall within the scope of the privilege. Second, the AI-generated reports at issue lack the requisite confidentiality. A fundamental requirement of the attorney-client privilege is that the communication be made in confidence between privileged parties. The defendant used a consumer-grade AI tool whose privacy policy expressly provides that user inputs may be used for model training and that the service provider reserves the right to disclose such data to regulatory authorities. Accordingly, such voluntary disclosure to a third party is generally treated as a waiver of any reasonable expectation of confidentiality. Third, communications between the defendant and the AI tool were not made for the purpose of obtaining legal advice. Claude expressly disclaims providing legal advice, and the defendant did not communicate with the AI tool at the direction of counsel. Accordingly, in the absence of instruction from counsel, communications between the defendant and the AI tool are not protected by the attorney-client privilege. This conclusion is not altered by the fact that the defendant subsequently shared the AI-generated reports with counsel.

Separately, communications between a party and an AI platform are not protected under the work product doctrine. The doctrine generally applies to materials prepared by an attorney or at the direction of counsel in anticipation of litigation. In this case, Bradley Heppner used the AI platform on his own initiative, without any direction from counsel. The court reasoned that the AI platform is neither an attorney nor an agent of counsel, and the materials it generated therefore cannot be treated as attorney work product.

United States of America v. Bradley Heppner marks an effort by the U.S. judiciary to draw a clear line regarding parties’ use of AI. Where parties process confidential information on consumer-grade AI platforms lacking data protection safeguards, they face a significant risk that their communications with such platforms may be used against them as evidence.

II.The Boundaries of Lawyers’ Use of AI Tools in Judicial Proceedings

Against the backdrop that parties’ use of AI in judicial proceedings typically requires the direction of counsel, the question arises as to how the boundaries of lawyers’ own use of AI tools should be defined. In recent years, various jurisdictions have begun to issue guidance on the use of AI by lawyers. For example, in late 2025, the Shanghai Bar Association released the Guidelines on the Use of AI Tools in Providing Community Legal Advisory Services (2025) (Trial), aimed at enhancing lawyers’ competence in using AI tools and preventing significant risks. In March 2026, Singapore issued the Guide for Using Generative AI in the Legal Sector⁹. These guidelines consistently emphasize that lawyers must ensure data security when using AI tools, remain responsible for the final work product, and exercise caution regarding the risks of so-called “AI hallucinations”.

AI hallucinations have become a major source of risk, and may even give rise to litigation, in connection with lawyers’ use of AI tools in recent years. In Mata v. Avianca¹⁰, counsel submitted six fictitious cases generated by ChatGPT. The court found that, after the opposing party raised challenges, counsel not only failed to withdraw the submissions but continued to rely on AI-generated false explanations, thereby acting in subjective bad faith. The court ultimately imposed a sanction of USD 5,000. This case illustrates that the use of AI tools itself does not constitute misconduct; however, submitting unverified false information and attempting to conceal it may amount to an abuse of the judicial process.

Similar issues have also arisen in Chinese judicial practice. In April 2025, the People’s Court of Tongzhou District of Beijing adjudicated a dispute involving Ren against Cao and a Beijing International Travel Agency Co., Ltd., which was subsequently included in the national case database of the People’s Courts of Tongzhou District of Beijing¹¹. In that case, the cases cited in counsel’s submissions were not genuine judicial precedents, but were instead generated through repeated prompting of an AI language model using keywords and disputed issues derived from the facts of the case.

In this case, the People’s Court of Tongzhou District of Beijing noted that Article 114(1) of the PRC Civil Procedure Law provides: “Where a litigation participant or any other person commits any of the following conduct, the people’s court may impose a fine or detention on the litigation participant or person according to the severity of the circumstances; and if suspected of any crime, the litigation participant or person shall be subject to criminal liability in accordance with law. (1) Forging or destroying any material evidence, which obstructs the trial of the case by the people's court.” The court further clarified that the “evidence” referred to in this provision should be understood as evidence material to the determination of the facts of the case. Cases submitted by litigation participants for reference purposes fall within the scope of legal arguments rather than evidence. Accordingly, the submission of unverified false cases does not constitute “forging or destroying any material evidence” within the meaning of this provision. However, the submission of false cases violates the principle of good faith in litigation and disrupts the order of civil proceedings, and the responsible party may bear corresponding legal consequences depending on the circumstances. In practice, the people’s courts may, in light of factors such as the degree of subjective fault, the harmful consequences, and the extent of disruption to the proceedings, impose sanctions by reference to Article 114(1) of the PRC Civil Procedure Law.

Therefore, existing guidelines and cases indicate that the boundaries governing lawyers’ use of AI tools in judicial proceedings include, but are not limited to, the following: First, when using AI tools, lawyers remain subject to confidentiality obligations consistent with professional conduct rules and must not disclose client information. Second, while AI tools may be used as auxiliary tools, lawyers bear an independent duty to review and verify the final work product they submit and may not rely on AI-generated content to discharge their own duty of care. Third, the submission of false or fabricated materials constitutes a violation of the duty of candor in litigation. Depending on the circumstances, such conduct may result in consequences ranging from judicial admonishment to findings of abuse of the judicial process and the imposition of sanctions, including monetary penalties.

III.Practical Implications

The use of AI has the potential both to assist parties in addressing legal issues, thereby lowering barriers to legal services and reducing information asymmetries, and to enhance lawyers’ efficiency in handling cases. It is therefore foreseeable that, as AI technologies become more widely adopted in legal practice, procedural disputes and disciplinary actions arising from the use of AI will continue to increase. In response, judicial authorities and professional regulators are likely to move from case-by-case exploration toward the development of clearer and more comprehensive rules governing the boundaries of AI use.

Against this backdrop, while lawyers may use AI tools subject to professional duties and regulatory constraints, parties acting without legal guidance lack such safeguards, thereby facing significantly greater risks.

The court’s ruling in Heppner demonstrates that when a party independently engages with a generative AI platform to discuss legal issues, the resulting communications will generally not be protected by either the attorney-client privilege or the work product doctrine, and may therefore be obtained and reviewed by the government, regulatory authorities, or courts as evidence.

By contrast, confidential communications between a party and counsel made for the purpose of obtaining legal advice are typically protected by the attorney-client privilege. Likewise, legal analyses, litigation strategies, and other materials prepared by counsel in connection with a case are protected under the work product doctrine, and thus are generally exempt from disclosure.

Beyond the applicability of privilege and related protection doctrines, courts may also directly restrict the use of AI platforms in handling case-related information. In Sohyon Warner v. Gilbarco, Inc.¹², during the discovery phase, the defendant expressed concerns about producing certain documents out of fear that such materials might be uploaded to AI platforms. In response, the United States District Court for the Eastern District of Michigan, Southern Division, modified the existing protective order to expressly provide that “any document marked confidential shall not be uploaded onto any AI platform.”¹³ This ruling indicates that, in the context of discovery, courts may, through protective orders or other mandatory directives, expressly prohibit parties from uploading confidential materials obtained through the discovery process to AI platforms, thereby restricting the use of such platforms for processing or analyzing case-related information.

In light of the foregoing, parties handling litigation matters in common law jurisdictions should refrain from independently using generative AI platforms to discuss case-related issues, and in particular should avoid inputting core facts, legal analyses, or potential litigation strategies. To maximize the client’s interests in litigation, it is advisable that litigation matters be handled comprehensively by qualified counsel.

Footnotes

1 Federal Rules of Civil Procedure

Rule 26. Duty to Disclose; General Provisions Governing Discovery

(b) Discovery Scope and Limits.

(1) Scope in General. Unless otherwise limited by court order, the scope of discovery is as follows: Parties may obtain discovery regarding any nonprivileged matter that is relevant to any party’s claim or defense and proportional to the needs of the case, considering the importance of the issues at stake in the action, the amount in controversy, the parties’ relative access to relevant information, the parties’ resources, the importance of the discovery in resolving the issues, and whether the burden or expense of the proposed discovery outweighs its likely benefit. Information within this scope of discovery need not be admissible in evidence to be discoverable.

2 Federal Rules of Criminal Procedure

Rule 16. Discovery and Inspection

(b) Defendant’s Disclosure.

(1) Information Subject to Disclosure.

(A) Documents and Objects. If a defendant requests disclosure under Rule 16(a)(1)(E) and the government complies, then the defendant must permit the government, upon request, to inspect and to copy or photograph books, papers, documents, data, photographs, tangible objects, buildings or places, or copies or portions of any of these items if:

(i) the item is within the defendant’s possession, custody, or control; and

(ii) the defendant intends to use the item in the defendant’s case-in-chief at trial.

3 Federal Rules of Evidence

Rule 502. Attorney-Client Privilege and Work Product; Limitations on Waiver

(g) Definitions. In this rule:

(1) “attorney-client privilege” means the protection that applicable law provides for confidential attorney-client communications; and

(2) “work-product protection” means the protection that applicable law provides for tangible material (or its intangible equivalent) prepared in anticipation of litigation or for trial.

4 Federal Rules of Evidence

Rule 501. Privilege in General

The common law — as interpreted by United States courts in the light of reason and experience — governs a claim of privilege unless any of the following provides otherwise:

·the United States Constitution;

·a federal statute; or

·rules prescribed by the Supreme Court.

But in a civil case, state law governs privilege regarding a claim or defense for which state law supplies the rule of decision.

5 Consolidated Laws of New York

Civil Practice Law & Rules

ARTICLE 31 Disclosure

§ 3101. Scope of disclosure.

(b) Privileged matter. Upon objection by a person entitled to assert the privilege, privileged matter shall not be obtainable.

(c) Attorney’s work product. The work product of an attorney shall not be obtainable.

6 Texas Rules of Evidence

Rule 503. Lawyer–Client Privilege

(b) Rules of Privilege.

(1) General Rule. A client has a privilege to refuse to disclose and to prevent any other person from disclosing confidential communications made to facilitate the rendition of professional legal services to the client:

(A) between the client or the client’s representative and the client’s lawyer or the lawyer’s representative;

(B) between the client’s lawyer and the lawyer’s representative;

(C) by the client, the client’s representative, the client’s lawyer, or the lawyer’s representative to a lawyer representing another party in a pending action or that lawyer’s representative, if the communications concern a matter of common interest in the pending action;

(D) between the client’s representatives or between the client and the client’s representative; or

(E) among lawyers and their representatives representing the same client.

TEXAS RULES OF CIVIL PROCEDURE

RULE 192. PERMISSIBLE DISCOVERY: FORMS AND SCOPE; WORK PRODUCT; PROTECTIVE ORDERS; DEFINITIONS

192.5 Work Product.

(b) Protection of work product.

(1) Protection of core work product--attorney mental processes. Core work product - the work product of an attorney or an attorney’s representative that contains the attorney’s or the attorney’s representative’s mental impressions, opinions, conclusions, or legal theories - is not discoverable.

7 HM Revenue & Customs. ECSH110250 - Legal Professional Privilege. Economic Crime Supervision Handbook. GOV·UK(https://www.gov.uk/hmrc-internal-manuals/economic-crime-supervision-handbook/ecsh110250)

8 UNITED STATES OF AMERICA v BRADLEY HEPPNER, 25 Cr. 503 (JSR), MEMORANDUM, Dkt. 27 (S.D.N.Y. Feb. 17, 2026)

9 Guide for Using Generative AI in the Legal Sector, published on 6 March 2026, via:

https://www.mlaw.gov.sg/files/Guide_for_using_Generative_AI_in_the_Legal_Sector__Published_on_6_Mar_2026_.pdf

10 Mata v. Avianca, Inc., No. 1:22-cv-01461, (S.D.N.Y.)

11 The case is docketed as (2024) Jing 0112 Min Chu No. 19067, and its reference number in the People’s Court Case Database is 2025-18-2-494-001.

12 SOHYON WARNER v. GILBARCO, INC., GILBARCO, INC. (d/b/a GILBARCO VEEDER-ROOT) and VONTIER CORPORATION, Case No. 2:24-cv-12333

13 SOHYON WARNER v. GILBARCO, INC., GILBARCO, INC. (d/b/a GILBARCO VEEDER-ROOT) and VONTIER CORPORATION, Case No. 2:24-cv-12333, ECF No. 66 (E.D. Mich. S. Div. Oct. 30, 2025), at 3

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.