While courts have extended work-product protection to AI used in preparing litigation, a very different situation arises when AI is consulted during a deposition. A federal court in Michigan ruled in April 2026 that using ChatGPT to answer deposition questions in real time is neither permissible nor protectable — and that a pro se plaintiff’s attempt to claim attorney-client privilege over AI interactions, while having no attorney, was not only legally wrong but spectacularly misconceived.
Background and Facts
Littiece Jones brought suit against Delta Air Lines, Inc. in the Eastern District of Michigan, proceeding as a pro se plaintiff. The litigation reached the deposition stage, and it was during her deposition that events took an unusual turn. Opposing counsel or, possibly, the court reporter observed that Jones had ChatGPT open and was actively consulting it while answering deposition questions. She was, in effect, using the AI tool as a real-time assistant to formulate or refine her answers during the examination under oath.
When the AI use was raised during the deposition, Jones refused to answer questions about it. Her stated basis for this refusal was attorney-client privilege. She declined to characterise her interactions with ChatGPT or to address the nature or extent of her reliance on it during the deposit on, citing a privilege that protects communications between a client and a licensed attorney — a privilege that requires, as a foundational element, the existence of an attorney-client relationship.
Littiece Jones had no attorney. She was a self-represented litigant. She did not have, and had not had, an attorney-client relationship with any lawyer in connection with this litigation. The privilege she invoked was therefore inapplicable on its own terms: there was no attorney and accordingly no attorney-client relationship from which a privilege could arise.
The matter came before the court, and the ruling that followed addressed both the privilege claim and the conduct that had prompted it.
What the Court Decided and Why
The court’s disposition of the attorney-client privilege claim was swift and categorical. Attorney-client privilege protects confidential communications made by a client to an attorney for the purpose of obtaining legal advice. The privilege is not available to communications between a person and a software application. ChatGPT is not an attorney. It is not licensed. It does not form an attorney-client relationship. It does not owe duties of confidentiality under professional conduct rules. The invocation of attorney-client privilege to shield interactions with an AI tool was, in the court’s analysis, simply inapplicable — the privilege did not arise and could not be asserted.
The court’s second ruling addressed the prospective use of AI during depositions. The court ordered that Jones could not use ChatGPT, or any other AI tool or any person, to assist her in answering deposition questions going forward. This order treats AI-assisted deposition answering as equivalent to — and prohibited alongside — having a person whisper answers to a deponent. A deposition examines the witness’s knowledge, recollection, and judgment. The answers must come from the witness. Consulting an external source in real time during a deposition, whether that source is a human advisor or an AI system, undermines the fundamental purpose of the deposition and the reliability of the testimony it is designed to elicit.
There is an important distinction embedded in this ruling that is worth making explicit. The cases discussed elsewhere in this series — Warner, Morgan, Tym — establish that preparing litigation materials with AI assistance is protected. Preparing for a deposition using AI, studying anticipated questions, drafting preparation notes with AI assistance — these preparatory activities are, on those cases’ reasoning, within the protected space. What Jones v. Delta Air Lines prohibits is something qualitatively different: consulting AI in real time during the examination itself, to formulate or refine answers while under oath. The deposition captures the witness’s own testimony, not the output of a real-time AI query.
The Principle Established
The Jones ruling establishes two propositions of clear practical importance. First, attorney-client privilege does not and cannot attach to interactions with AI tools. This is not a surprising conclusion — the privilege has always required a human attorney as one of its necessary parties — but the explicit judicial statement is useful at a time when popular misunderstanding of what AI can and cannot do in a legal context is widespread.
Second, the use of AI during a deposition to assist in answering questions is not permissible and will be prohibited by court order. The deposition process depends on the candid and unassisted testimony of the witness. A witness who consults an AI tool in real time is not giving their own testimony; they are giving a hybrid of their recollection and whatever the AI happens to generate in response to a prompt framed by the witness’s understanding of the question. This is not a permissible form of testimony, and courts will intervene to prevent it.
The case also serves as a cautionary example for pro se litigants more generally. The expansion of AI tools has empowered self-represented parties to navigate complex litigation in ways that were not previously possible. But that empowerment does not override foundational procedural obligations, and it does not create privileges that do not exist. Pro se litigants who use AI must understand the distinction between permissible preparation and impermissible real-time assistance during examination under oath.
The India Angle
The Jones ruling is directly relevant to the conduct of examination-in-chief and cross-examination in Indian courts, even though the procedural context differs. Under the Indian Evidence Act, 1872, and under the Code of Civil Procedure, 1908 and the Code of Criminal Procedure, 1973, examination of witnesses is intended to elicit the witness’s own knowledge and recollection. The question of whether a witness who consults an AI tool in real time during examination is giving permissible testimony has not yet arisen explicitly in Indian reported case law, but the principle is not in doubt: testimony must come from the witness, not from an AI query run on the witness’s behalf.
The privilege dimension of Jones also has a clear Indian parallel. Indian law recognises professional communications privilege under Sections 126 and 129 of the Evidence Act. These provisions protect communications between an advocate and client. The question of whether an Indian litigant could assert equivalent protection over their interactions with an AI tool — perhaps arguing that the AI was functioning as a legal advisor — is not one that has been tested, but the answer, applying the reasoning of the Jones court, is straightforwardly no. Privilege under Indian law, as under American law, requires a human professional at one end of the communication. There is no basis in current Indian law for extending statutory privilege to AI interactions, and advocates should be cautious about advising clients that any such protection exists.
Case Details
| Case Name | Littiece Jones v. Delta Air Lines, Inc. |
| Court | United States District Court, Eastern District of Michigan |
| Decision Date | 22 April 2026 |
| Key Facts | Pro se plaintiff used ChatGPT in real time during deposition; refused to answer questions about this, claiming attorney-client privilege; had no attorney |
| Holding | Attorney-client privilege inapplicable (no attorney-client relationship with AI); plaintiff ordered not to use ChatGPT or any AI or person to assist in answering deposition questions |
| Distinction from Warner/Morgan/Tym | Prohibited conduct is real-time AI consultation during examination; preparatory AI use before deposition remains within protected space |
| Significance | First ruling prohibiting AI-assisted deposition answering; confirms attorney-client privilege cannot attach to AI tool interactions |