The Risks of Using Artificial Intelligence to Draft Wills and Testamentary Devices in New Jersey and New York

I. Introduction

Artificial intelligence (“AI”) continues its precipitous rise into everyday life, and many individuals are now relying on it to draft testamentary devices, such as wills and testamentary trusts, without consulting an attorney during the process. While AI-generated estate planning documents are not categorically prohibited in New Jersey or New York, and while they may appear polished and professionally drafted, they may harbor serious defects that become apparent only after death, when it is too late to remedy them. If you are considering using AI to draft your will, you should understand the significant risks involved. The potential cost savings and ease of use associated with AI systems like large language models (“LLMs”) must be measured against well-established legal requirements governing capacity, execution, and testamentary intent, each of which AI-assisted drafting can undermine in ways that may not surface until probate.

This article considers the major pitfalls that individuals face when using AI to draft their own wills. These include the failure to satisfy execution and attestation formalities required by law, the failure to accurately capture testamentary intent, and problems of ambiguity, misdescription, and interpretation that can lead to costly litigation among heirs. Additionally, AI-assisted drafting raises concerns about the loss of critical contemporaneous documentation regarding mental capacity and freedom from undue influence, evidence that could prove essential if someone later challenges the will in probate court.

II. Execution and Attestation Formalities: The Ceremony AI Cannot Supervise

AI can produce a seemingly professional document, but it cannot supervise the execution ceremony, and execution is where most wills fail. In New Jersey, the statutory requirements are straightforward: a will must be in writing, signed by the testator (or some other individual in the testator’s conscious presence and at the testator’s direction), and signed by at least two witnesses within a reasonable time after the will was signed (or signed by another at the testator’s direction). N.J.S.A. 3B:3-2(a)(1-3). New Jersey also recognizes holographic wills, which is an exception to N.J.S.A. 3B:3-2(a) where a will may be valid even without witnesses if the signature and material portions of the document are in the testator’s handwriting. N.J.S.A. 3B:3-2(b). This exception is, however, inapplicable to AI-generated documents, which are not handwritten. Furthermore, the New Jersey statute also allows a document that does not strictly comply with these formalities to be admitted to probate if clear and convincing extrinsic evidence is provided that shows the decedent intended their writing to be their last will and testament. N.J.S.A. § 3B:3-3. There is limited application of this provision to documents created by LLMs like ChatGPT, however. While one might argue that AI prompts, relevant web searches, and other digital records could be offered as extrinsic evidence of testamentary intent, the absence of contemporaneous observations about the testator’s understanding and intent may make that burden difficult to satisfy.

New York’s execution requirements are similar to New Jersey’s but contain an additional critical caveat: the will must be signed at the end of the document. Under EPTL § 3-2.1(a), every will must be in writing, signed by the testator at the end thereof, and attested by at least two witnesses. EPTL § 3-2.1(a)(1). The signature must be affixed in the presence of each attesting witness, or acknowledged by the testator to each of them. EPTL § 3-2.1(a)(2). The testator must declare to each witness that the instrument is his or her will. EPTL § 3-2.1(a)(3). The witnesses must, within a single 30-day period, both attest the testator’s signature and sign their names at the end of the will. EPTL § 3-2.1(a)(4). Critically, any text following the testator’s signature (other than the attestation clause) is not given any effect. EPTL § 3-2.1(a)(1)(B). If the AI places dispositive provisions after the signature line, those provisions will be disregarded entirely. Moreover, New York draws a stricter line with respect to extrinsic evidence, and only considers extrinsic evidence in the event there is a true ambiguity with what is written, not with respect to the aforementioned drafting requirements. Matter of Bieley, 91 N.Y.2d 520, n. 2 (1998).

The practical risk is significant. AI may fail to satisfy any of these requirements, rendering the Will invalid in its entirety. Anyone using AI to draft a will should understand that the technology delivers only a document, it cannot supervise the execution ceremony, ensure the testator declares the will to each witness, or confirm that witnesses sign within the applicable window.

III. Failure to Capture Testamentary Intent: When AI Gets It Wrong

A will must accurately capture testamentary intent, which is what the particular testator wants to happen with his or her property. AI cannot ascertain intent through dialogue, ask clarifying questions, detect subtle family dynamics, or investigate suspected errors. If AI produces language that does not reflect your true intentions and you go ahead and sign it anyway, the consequences depend heavily on which state’s law applies.

In New Jersey, courts take a liberal approach to correcting mistakes: the function of a court in construing a will is to ascertain and give effect to the “probable intention of the testator,” and extrinsic evidence of surrounding circumstances “should be admitted” for that purpose. Wilson v. Flowers, 58 N.J. 250, 260 (1971). In a recent decision, the New Jersey Appellate Court even found that the court may reform wills for mistake upon clear-and-convincing proof. Matter of Estate of Canova, No. A-1696-23, 2026 WL 305263 (N.J. Super. Ct. App. Div., Feb. 5, 2026). While drafting errors may be susceptible to judicial correction, the burden of proof remains substantial.

New York takes a far more rigid approach, reasoning that the “best indicator of the testator’s intent is found in the clear and unambiguous language of the will itself,” and where no ambiguity exists, [e]xtrinsic evidence is inadmissible to vary the terms of a will.” In re Estate of Scale, 38 A.D.3d 983, 985 (3d Dep’t 2007). This application creates a dangerous trap for AI-generated wills: a will that is “facially clear but wrong” may lock in an erroneous disposition with no judicial remedy. If the AI produces language that unambiguously directs property to the wrong person or entity or in the wrong shares, New York courts may very well enforce that language, even if it contradicts what the heirs believe the testator actually wanted.

AI-generated wills are prone to a range of errors that an experienced attorney would catch. Common mistakes include misdescription of property or beneficiaries, conflicting definitions, omitted residuary clauses, provisions based on the wrong jurisdiction’s law, outdated tax provisions, inconsistency with trusts or beneficiary designations, conflicts with marital agreements, and internally contradictory clauses. Unlike commercial contracts, wills are typically not reviewed by any potentially adverse parties before execution, which means these errors may remain undetected until the testator’s death, at which point the only remedy is costly and uncertain litigation among the heirs and beneficiaries.

 IV. Testamentary Capacity and Undue Influence: The Documentation Gap

A critical and underappreciated litigation risk of AI-assisted will drafting is the loss of contemporaneous documentation of the testator’s mental state.

In New Jersey, testamentary capacity is gauged by the testator’s ability to comprehend the property, know the natural objects of bounty, understand the nature of the testamentary act, and understand the distribution being made; only “a very low degree of mental capacity” is required, and the challenger bears the burden of establishing incapacity by clear and convincing evidence. Matter of Will of Liebl, 617 A.2d 266, 268 (App. Div. 1992). In New York, by contrast, the proponent of the will bears the burden of proving capacity under the three-factor Kumstar test: that the testator (1) understood the nature and consequences of execution; (2) knew the nature and extent of property; and (3) knew the natural objects of bounty and their relationships. Matter of Kumstar, 66 NY2d 691, 692 (1985), rearg denied 67 NY2d 647 (1986). Old age, physical infirmity, and even low-level mental degradation are not necessarily inconsistent with capacity, as the inquiry is whether the testator was lucid and rational at the time of execution.

When an attorney drafts a will, that attorney can later testify about the testator’s capacity including observations about demeanor, lucidity, and understanding of the document’s contents. Furthermore, “when an attorney drafts a will and supervises its execution, a presumption of regularity is raised that the will was properly executed.” In re Estate of Leach, 3 A.D.3d 763, 764 (3d Dep’t 2004). In these cases, the drafting attorney’s testimony regarding the testator’s capacity, and the attorney’s routine practice of ascertaining information about family, assets, and desired disposition, was key evidence supporting admission of the will to probate. When a will is drafted primarily through an AI interface, however, there is no direct interaction with an attorney during the drafting process, creating a significant evidentiary gap. If your will is later contested on grounds of incapacity or undue influence, there may be no one who can testify from personal observation about your lucidity, demeanor, or understanding of the document’s contents. AI cannot observe demeanor, assess cognitive function in real time, or document the circumstances of the drafting session in a manner admissible as evidence of capacity.

Undue influence presents a parallel documentation challenge. In New Jersey, undue influence requires a showing of (i) the existence of a confidential relationship between the testator and the person alleged to have exerted the undue influence; and (ii) suspicious circumstances as to the disputed will. See Haynes v. First Nat’l State Bank, 87 N.J. 163, 176-77 (1981). Undue influence is also defined as a kind of “mental, moral, or physical exertion of a kind and quality that destroys the free will of the testator” exerted by another individual. In Re Estate of Stockdale, 196 N.J. 275, 302-303 (2008). Ordinarily, the contestant bears the burden, but where the will benefits someone in a confidential relationship with the testator and additional “suspicious” circumstances exist, that “need only be slight,” the burden shifts to the proponent by clear and convincing evidence. Id.

In New York, undue influence must be proved by a showing of: (i) motive, (ii) opportunity, and (iii) the actual exercise of that undue influence.” Matter of Estate of Linich, 213 A.D.3d 1, 5 (3d. Dep’t 2023). “Where there is a confidential relationship between parties to a transaction, the burden shifts to the stronger party in such a relationship to prove by clear and convincing evidence that a transaction from which he or she benefitted was not occasioned by undue influence.” Matter of Mary, 202 A.D.3d 1418, 1420 (3d Dept. 2022) (internal quotation marks and citations omitted).

LLMs are fundamentally incapable of detecting undue influence. They cannot observe who is actually present during the drafting session, identify whether a third party is supplying the prompts or selecting beneficiaries, or perceive the interpersonal dynamics that might indicate coercion or manipulation. AI cannot assess whether the testator is acting of his or her own free will, detect the “slight” suspicious circumstances that would shift the burden under New Jersey law, or gather the evidence of motive, opportunity, and actual exercise of influence required in New York. These are inherently human, fact-bound judgments that must be made at the moment of execution by someone who can observe the testator’s demeanor, ask probing questions, and document the absence of suspicious circumstances. An AI interface simply cannot perform this essential gatekeeping function.

V. The AI Hallucination Problem: Lessons from Sanctions Cases

No jurisdiction has yet issued a reported decision involving an AI-drafted will, but courts have increasingly sanctioned attorneys for filing AI-generated documents containing fabricated citations—so-called “hallucinations.” In the estate planning context, these errors might manifest as nonexistent trust provisions, outdated tax exemptions, incorrect jurisdictional law, or beneficiary designations that conflict with the testator’s actual family structure.

Attorneys in both New Jersey and New York are bound by Rules of Professional Conduct that require competent representation, including the legal knowledge, skill, thoroughness, and preparation reasonably necessary to represent the client. N.J. RPC 1.1; N.Y. Rule 1.1. The requirement of competence includes keeping abreast of changes in technology. With respect to AI, this means attorneys have an ethical duty to verify all AI-generated content for accuracy, understand how the technology works and its limitations, and ensure that AI outputs are critically reviewed before use. N.J. RPC 3.1, 4.1(a)(1), 8.4(c); N.Y. Rule 3.3. Failure to comply with these obligations exposes attorneys to disciplinary action, including sanctions, suspension, or disbarment.

Ultimately, ensuring the accuracy and validity of a will is the attorney’s responsibility, not the LLM. While attorneys bear the ethical obligations set forth above, AI providers have no ethical obligations to clients and no accountability for errors in the content they generate.

VI. Confidentiality and Data Privacy Risks Specific to Estate Planning

Before using any AI tool for estate planning, you should understand what the platform does with the information you enter and whether adequate safeguards exist to protect your privacy. Artificial Intelligence platforms may also prompt (or require) you to enter highly sensitive personal information into AI systems, including family relationships, assets, liabilities, health information, beneficiary disputes, tax information, incapacity concerns, and disinheritance rationale. Many commercial AI platforms retain user inputs for model training and other purposes, creating a risk that this confidential information may be disclosed to or accessed by third parties. Attorneys are bound by strict confidentiality obligations that require reasonable efforts to prevent the unauthorized disclosure of client information. N.J. RPC 1.6; N.Y. Rule 1.6. AI providers have no such obligations and may use or share your data in ways you do not expect.

VII. Best Practices and Risk Mitigation Strategies

Because AI is not perfect, attorney involvement remains essential to ensuring that a will is valid, enforceable, and defensible against challenge. AI cannot supervise the execution ceremony or ensure compliance with attestation formalities. It cannot capture testamentary intent through the kind of dialogue that reveals family dynamics, tax objectives, and coordination with other estate planning instruments. It cannot assess whether you possess testamentary capacity or detect whether someone is exerting undue influence over you. It cannot verify that the provisions it generates are legally accurate or consistent with the law of your jurisdiction. And it cannot testify on your behalf if the will is later contested. For all of these reasons, if you are considering using AI to draft your will, you should consult with an experienced estate planning attorney who can review the document, ensure proper execution, and create the documentation that may be needed to defend your testamentary wishes in court.