Generative artificial intelligence is rapidly becoming part of the ordinary workflow of lawyers, clients, consultants, investigators, and self-represented litigants. Attorneys may use generative AI to organize timelines, analyze allegations, identify factual inconsistencies, generate research queries, test legal theories, prepare deposition outlines, summarize discovery, or produce an initial draft of a pleading or brief. Clients may use the same technology to understand a dispute, prepare information for counsel, or evaluate possible claims and defenses. These uses create records that did not previously exist in traditional litigation files, including prompts, uploaded documents, generated responses, chat histories, revision requests, account metadata, and system logs.
The emergence of these records has produced an important discovery question: When are communications with a generative AI system, and the materials generated through that system, protected by the attorney work-product doctrine?
The answer is developing quickly. Courts have not adopted a single rule that all AI prompts and outputs are protected, nor have they concluded that using a third-party AI platform automatically waives work-product protection. Instead, the emerging decisions generally apply traditional work-product principles to the circumstances surrounding the particular AI use. The central questions remain why the material was created, whether litigation was anticipated, who created or directed the work, what the material reveals about litigation strategy, and whether the user preserved sufficient confidentiality.
For lawyers and businesses, the practical lesson is that generative AI is not a protected category unto itself. An AI-generated document may be protected for the same reasons that a handwritten trial outline, attorney memorandum, investigator’s report, or draft pleading would be protected. Conversely, an AI conversation may be discoverable when it was created for an ordinary business purpose, generated independently by a client without counsel’s direction, shared broadly, or stored under conditions inconsistent with maintaining confidentiality.
The phrase “work-product privilege” is commonly used, but “work-product protection” or “work-product doctrine” is usually more precise. The attorney-client privilege protects confidential communications made for the purpose of requesting or providing legal advice. The work-product doctrine protects qualifying materials prepared because of actual or anticipated litigation. The doctrines frequently overlap, but they protect different interests and are subject to different waiver principles.
The modern doctrine originated in Hickman v. Taylor, in which the United States Supreme Court recognized that attorneys must be able to prepare cases without unnecessary intrusion by opposing counsel into their files, impressions, interviews, legal theories, and strategic judgments. ¹ Federal Rule of Civil Procedure 26(b)(3) later incorporated that protection into the federal discovery rules. It ordinarily shields documents and tangible things prepared in anticipation of litigation or for trial by or for a party or its representative, including the party’s attorney, consultant, insurer, indemnitor, or agent. ²
Rule 26 distinguishes between ordinary or fact work product and opinion work product. Ordinary work product may contain factual information gathered or organized for litigation. It can sometimes be discovered if the requesting party demonstrates a substantial need for the information and cannot obtain its substantial equivalent without undue hardship. Opinion work product receives significantly stronger protection because it reflects an attorney’s mental impressions, conclusions, opinions, or legal theories. Even when a court orders the production of qualifying factual work product, Rule 26 directs the court to protect those mental impressions and legal theories from disclosure. ²
Generative AI materials do not fit perfectly into the traditional terminology of “documents and tangible things,” but that does not place them outside the doctrine. A saved prompt, exported chat, generated memorandum, revision history, or electronic transcript is a document or form of electronically stored information. More importantly, an AI conversation may reveal the user’s thought process in considerable detail. A sequence of prompts may show which facts counsel considered significant, which witnesses counsel doubted, which legal theories were being evaluated, what weaknesses counsel perceived, and how counsel intended to frame the dispute.
In some circumstances, the prompts may reveal more about an attorney’s strategy than the final document. A finished motion may contain only the legal position that counsel elected to advance. The underlying AI conversation may reveal alternative theories that were rejected, unfavorable facts counsel was attempting to address, anticipated counterarguments, assessments of witness credibility, and confidential instructions supplied by the client. That content may constitute classic opinion work product even if the AI-generated final draft is later revised substantially or never used.
The most important work-product question is ordinarily whether the material was prepared in anticipation of litigation. The fact that a lawyer used an AI system does not establish that requirement. Lawyers use AI for litigation, but they may also use it to write marketing materials, prepare routine correspondence, summarize public information, negotiate ordinary contracts, conduct general compliance reviews, or perform administrative work. Those uses do not become work product merely because an attorney performed them.
Courts generally examine whether the material was created because of the prospect of litigation rather than in the ordinary course of business. A company cannot necessarily shield an ordinary business analysis by sending it to counsel or entering it into an AI tool after a dispute develops. Similarly, a routine internal report does not automatically become work product because litigation was one possible consequence of the underlying event.
The protection becomes stronger when the AI activity is tied to an identifiable claim, investigation, demand, lawsuit, or reasonably anticipated dispute. An attorney who uploads a complaint and asks an approved AI system to identify possible affirmative defenses is using the system for a litigation purpose. An attorney who instructs the system to compare deposition testimony with contemporaneous emails, generate a cross-examination outline, or identify evidentiary weaknesses is likewise creating material because of litigation. When those prompts reflect counsel’s strategic choices, they may contain opinion work product of the highest order.
Timing is relevant but not conclusive. Materials created after a complaint is filed are not automatically protected, and materials created before filing are not automatically unprotected. A pre-suit AI analysis may qualify when the dispute has become sufficiently concrete, and the analysis was conducted to prepare for likely litigation. Conversely, an analysis generated during pending litigation may remain discoverable if it was produced for an ordinary commercial or operational purpose unrelated to the prosecution or defense of the case.
Documentation can make a difference. Matter files should show that counsel authorized the AI use for a particular litigation purpose. The file need not contain artificial labels designed merely to manufacture protection. It should, however, provide an accurate record of the purpose for which the material was created, the individuals involved, the matter to which it relates, and the role of counsel in directing or supervising the work.
Rule 26 does not protect only documents personally written by an attorney. It extends to qualifying materials prepared by or for the party or its representative.² Therefore, an attorney’s use of AI is not the only situation in which protection may apply. A paralegal, investigator, litigation consultant, retained expert, insurer, or other representative may use an AI system at counsel’s direction to prepare materials for litigation.
Client-generated AI material presents a more difficult issue. A client may use AI after counsel instructs the client to organize facts, reconstruct a timeline, identify relevant communications, or prepare an account of anticipated testimony. When the client’s AI use is undertaken at counsel’s request and for the purpose of assisting counsel with litigation, a strong argument exists that the resulting prompts and outputs were prepared by or for the party in anticipation of litigation.
The argument is weaker when a client independently consults a public AI chatbot before retaining counsel, or without counsel’s knowledge or direction. Such a conversation might still reflect the client’s own preparation for litigation, particularly because Rule 26 protects qualifying material created by or for a party and does not invariably require attorney involvement. The absence of counsel’s direction, however, creates additional uncertainty about the purpose of the material, the expectation of confidentiality, and whether the conversation reflects legal strategy developed for the litigation.
A law firm should therefore address client use expressly. Counsel may instruct clients not to upload privileged communications, confidential discovery, medical information, trade secrets, personal identifying information, or other sensitive material to an unapproved public AI system. When client-assisted AI work would be useful, counsel should define the assignment, identify an approved platform, explain confidentiality requirements, and preserve a record that the work was performed at counsel’s direction for the litigation.
Several decisions issued in 2026 illustrate the developing judicial approaches. Although the factual circumstances and governing procedural rules differ, the decisions show that courts are focusing on purpose, direction, confidentiality, and the actual content of the AI conversations rather than applying a universal rule.
In United States v. Heppner, a federal district court considered whether a criminal defendant’s communications with the public version of Anthropic’s Claude platform were protected by the attorney-client privilege or work-product doctrine. The court concluded that the materials were not protected. It emphasized that the defendant had created the AI conversations on his own initiative rather than at counsel’s direction and that the documents did not reflect defense counsel’s strategy or legal theories. The court also questioned whether the communications were confidential in light of the platform’s policies and the potential for third-party access.⁶
Heppner represents a comparatively narrow view of protection for independently generated client-AI communications. The decision did not establish that attorneys themselves lose work-product protection whenever they use AI. Rather, the court examined the particular defendant’s conduct and found insufficient connection between the AI records and counsel’s preparation of the defense.⁶
Only days earlier, the United States District Court for the Eastern District of Michigan reached a different result in Warner v. Gilbarco, Inc. There, the court addressed discovery seeking information about a self-represented plaintiff’s use of generative AI. The court treated the plaintiff’s AI-assisted litigation materials as work product and rejected the proposition that entering information into an AI program necessarily waived protection. It reasoned that programs such as ChatGPT are tools rather than persons and applied the traditional work-product waiver inquiry, which ordinarily focuses on disclosure to an adversary or conduct that creates a substantial likelihood the material will reach an adversary.⁷
The Warner decision is particularly significant because it rejected a categorical equation between AI use and disclosure to an opposing party. Under that reasoning, using a technological tool does not itself destroy protection, just as using word-processing software, cloud storage, electronic legal research, or a litigation database does not automatically disclose work product to an adversary. The relevant analysis concerns how the tool operates, who may access the information, and whether the user took reasonable measures to preserve the adversarial confidentiality that the doctrine protects.⁷
In Morgan v. V2X, Inc., a federal district court in Colorado similarly recognized that a self-represented litigant’s AI-assisted preparation could contain protected mental impressions and litigation strategy. The court declined to hold that use of a publicly available AI platform automatically eliminated every reasonable expectation of privacy or waived all work-product protection. At the same time, it distinguished between protected substantive content and unprotected information about the technology itself. The court required disclosure of the identity of the AI platform that had been used with confidential materials and strengthened the protective order to restrict the uploading of protected discovery to mainstream generative AI tools without adequate safeguards.⁸
Morgan therefore reflects a functional compromise. The substance of an AI conversation may be protected, but the party may still have to disclose enough information about the system, account, or handling of confidential discovery to allow the court and opposing party to evaluate compliance with discovery obligations and protective orders.⁸
The Texas Business Court adopted another protection-oriented approach in Tate Group Automotive, LLC v. Legacy Automotive Capital, LLC. After reviewing the disputed ChatGPT conversations in camera, the court held that most of the principal’s AI conversations constituted work product under the Texas rule, which protects material and mental impressions developed in anticipation of litigation by or for a party or its representatives. The court followed the functional reasoning of Warner and Morgan and declined to adopt the broader implications of Heppner.⁹
The protection was not absolute. The court required production of portions that did not qualify as work product and directed disclosure concerning discovery materials that had been shared with ChatGPT. It also encouraged the parties to address AI use in their protective order. Tate Group demonstrates that a court may protect the strategic substance of AI chats while separately regulating the transfer of confidential discovery into the AI system.⁹
A New York state court reached a related conclusion in Assini v. Hayward. A subpoena to OpenAI sought prompts, inputs, uploads, outputs, and related information associated with the drafting of litigation materials. The court quashed the subpoena, reasoning that iterative communications with a generative AI platform may contain the type of confidential, strategy-laden material traditionally protected as work product. The court relied on Morgan and rejected a sweeping interpretation of Heppner. It nevertheless cautioned that AI use must comply with governing court rules and ethical duties and that improper use could lead to sanctions.¹⁰
These decisions do not create a settled national rule. They instead identify the factual questions that will likely control future disputes. Courts will examine whether the AI use was connected to actual or anticipated litigation, whether counsel directed or supervised it, whether the content reveals strategic mental impressions, what information was submitted to the platform, who could access it, how the provider could use it, and whether the user complied with protective orders and other confidentiality obligations.
Waiver of work-product protection is not identical to waiver of the attorney-client privilege. Attorney-client privilege generally depends on maintaining confidentiality from unnecessary third parties. Work-product protection is designed primarily to preserve the privacy of preparation against litigation adversaries. Many courts therefore focus on whether protected material was disclosed to an adversary or in a manner substantially likely to place it in an adversary’s hands.
That distinction supports the reasoning that using a third-party technology provider does not automatically waive work product. Law firms routinely rely on cloud-based document platforms, electronic discovery vendors, court-reporting companies, legal research systems, translators, experts, and other service providers. The mere existence of a third-party service provider cannot reasonably mean that every document processed through the service loses protection.
The analysis becomes more difficult when the provider’s terms permit broad use of submitted information, human review unrelated to providing the service, retention for model training, disclosure to numerous subcontractors, or public sharing. A user who places litigation strategy into a publicly accessible AI conversation or intentionally publishes the output may have difficulty maintaining that reasonable precautions were taken. The same may be true when the user employs a consumer account despite firm policies requiring a protected enterprise system.
Platform labels alone should not determine waiver. A system marketed as “enterprise” may still have unfavorable retention, access, or subcontractor terms. A consumer system may offer settings that prevent model training but still retain information for security or legal purposes. The correct inquiry requires review of the actual contractual terms, privacy policies, administrative controls, retention practices, technical safeguards, and account configuration in effect when the material was submitted.
Law firms should not assume that deleting a chat from the user interface eliminates every stored copy. Nor should they assume that an opt-out setting resolves all confidentiality concerns. Vendor diligence should address whether prompts and uploaded files are used to train models, how long data is retained, whether administrators or provider personnel may review content, where the information is stored, what subprocessors receive it, how legal demands are handled, and whether the provider offers enforceable deletion and security commitments.
Federal Rule of Evidence 502 may reduce the consequences of certain inadvertent disclosures. Under Rule 502, an inadvertent disclosure in a federal proceeding does not necessarily constitute waiver when the holder took reasonable steps to prevent disclosure and promptly took reasonable steps to correct the error.³ The rule supports carefully drafted clawback provisions and protective orders, but it does not replace reasonable AI governance. A party cannot safely rely on a clawback agreement as permission to upload protected materials indiscriminately into systems whose risks have never been evaluated.
Even when AI material is ultimately protected, it may fall within the scope of a discovery request. Prompts, outputs, chat titles, uploaded files, timestamps, exported transcripts, and system logs may constitute electronically stored information. A party that reasonably anticipates litigation should consider whether relevant AI records must be preserved, especially when the system has been used to reconstruct events, analyze evidence, or communicate factual information about the dispute.
Preservation does not mean that all AI records must be produced. It means that potentially relevant information should not be destroyed before privilege and work-product determinations can be made. A party that deletes an unfavorable AI conversation after receiving discovery requests may create a preservation dispute even if portions of the conversation could have been protected. The better practice is to preserve first and then determine whether the material is responsive, protected, irrelevant, or subject to a protective order.
When AI materials are withheld as work product, Rule 26(b)(5) requires the withholding party to expressly assert the protection and describe the nature of the material in a manner that allows the opposing party to assess the claim without revealing the protected information. ² A privilege or work-product log may therefore need to identify the date of the conversation, the creator, the general category of AI system, the litigation matter, the general subject, the reason the material was created, and whether it was prepared at counsel’s direction.
The description should be specific enough to demonstrate the litigation nexus but should not reveal the strategy being protected. An entry stating only “AI chat—work product” is unlikely to provide meaningful information. A more useful description might explain that the record contains counsel-directed analysis of deposition testimony prepared for cross-examination in pending litigation. Depending on the circumstances, a court may conduct an in camera review when the parties cannot resolve the issue from the log and surrounding evidence.
The existence of protected prompts does not necessarily protect every fact mentioned in them. Work product generally protects the documents and the manner in which counsel selected, organized, or analyzed information; it does not automatically immunize underlying facts from discovery. An opposing party may be entitled to ask a witness what happened, what records exist, or who possesses relevant information, even though it cannot compel production of counsel’s strategic AI conversation about those subjects.
Traditional protective orders frequently prohibit unauthorized disclosure of confidential information but say nothing about generative AI. That omission can create disagreement over whether uploading protected discovery into an AI system constitutes disclosure, whether the provider is an authorized service provider, and what safeguards must be in place.
An effective protective order should distinguish between approved AI uses and systems that create unacceptable risk. It may permit use of a secure system that does not train on matter data, provides contractual confidentiality, limits retention, controls administrative access, and binds subprocessors to comparable obligations. It may prohibit submission of protected discovery to public or consumer AI systems unless the producing party consents or the court authorizes the use.
The order should also address deletion, security incidents, subpoenas directed to the provider, access by foreign subprocessors, audit rights, and the handling of information when the matter ends. The goal is not necessarily to prohibit all AI use. A blanket prohibition may prevent legitimate efficiencies without materially improving security. The more sustainable approach is to establish objectively verifiable conditions for the use of AI with protected information.
Counsel should raise the subject early, particularly in cases involving source code, trade secrets, health records, financial information, employee data, proprietary research, or personal identifying information. Waiting until after a party has uploaded extensive discovery to an unapproved system may leave the court with only remedial options. The decisions in Morgan and Tate Group illustrate that courts may protect the strategic content of AI-assisted work while still imposing strict controls on the submission of confidential discovery to AI providers.⁸ ⁹
Work-product protection is only one part of responsible AI use. Lawyers remain subject to professional obligations of competence, confidentiality, communication, supervision, candor, and reasonable billing. The American Bar Association’s Formal Opinion 512 explains that attorneys using generative AI must understand the technology sufficiently to identify its benefits and risks, protect client information, supervise its use, verify its output, and communicate with clients when the circumstances require consultation or informed consent.⁴
Competence does not require every attorney to become a computer scientist. It does require lawyers to understand that generative AI can produce inaccurate statements, nonexistent authorities, distorted summaries, and incomplete analyses. Lawyers must review the underlying sources and exercise independent professional judgment before relying on AI-generated content. The use of AI does not transfer responsibility for a filing, discovery response, contract, or legal opinion to the technology provider.
Confidentiality requires more than removing the client’s name from a prompt. A combination of facts may identify the client or disclose sensitive strategy even when names are omitted. Uploaded documents may contain hidden metadata, personal information, trade secrets, or privileged communications. Before using an AI system, counsel should evaluate both the information being submitted and the system receiving it.
The State Bar of Michigan has similarly cautioned attorneys to verify AI-generated material, remain responsible for documents submitted to tribunals, and exercise extreme care before entering client information into an AI platform. Its guidance emphasizes the importance of evaluating a provider’s security and confidentiality protections and considering whether client consent is appropriate.⁵
Communication with the client may be necessary when the proposed use creates a material confidentiality risk, affects the method or cost of representation, or involves unusually sensitive information. A law firm’s general engagement agreement may address routine use of secure technology, but a broad technology clause should not be treated as universal consent to use any AI provider under any conditions.
Supervision duties apply when associates, paralegals, contract attorneys, vendors, or clients use AI in connection with the representation. A firm policy that merely says “use AI responsibly” provides little meaningful direction. Lawyers should know which systems are approved, which categories of information may be entered, what verification is required, how outputs should be stored, and when senior or specialized review is necessary.
The strongest work-product position begins before a discovery dispute arises. A firm should maintain a written AI policy that identifies approved platforms and describes the circumstances under which litigation materials may be processed. The policy should reflect actual technological and contractual diligence rather than relying on marketing claims.
Matter-level controls are equally important. Highly sensitive matters may require a closed or internally hosted model, while lower-risk matters may permit a secure commercial system. The classification should take into account protective orders, contractual confidentiality obligations, data-protection laws, client requirements, and the types of information involved.
Counsel should separate client matters within the AI environment. Shared chat histories and personal accounts increase the possibility of cross-matter disclosure, accidental reuse, and incomplete preservation. Enterprise workspaces should use role-based access, multifactor authentication, administrative controls, and procedures for departing personnel. Matter names and labels should be designed to support preservation without unnecessarily disclosing confidential information to unauthorized administrators.
The relationship between counsel and the person using the system should also be documented. When a client, investigator, or consultant is asked to use an approved AI tool, the instruction should explain the litigation purpose, the permitted information, and the expected deliverable. That record can establish that the material was prepared at counsel’s direction and because of the litigation rather than for personal curiosity or an ordinary business purpose.
Lawyers should preserve meaningful human authorship and supervision. Work-product protection should not depend on pretending that an AI output was written entirely by counsel. The defensible position is that counsel selected the relevant information, framed the questions, evaluated the responses, rejected inaccurate or unhelpful material, and incorporated the useful portions into counsel’s own analysis. Those activities constitute legal judgment, even when technology assists with the initial processing.
Retention policies should account for the possibility that prompts and outputs may become relevant. Automatically deleting every chat after a few days may create preservation problems once litigation is anticipated. Retaining every AI conversation forever may create unnecessary cost, risk, and discovery burdens. The firm should adopt a matter-based schedule that preserves necessary records while disposing of transitory or duplicative material when legally appropriate.
Before producing AI-related material, counsel should determine whether the responsive information can be separated from protected analysis. Some portions may contain only nonprivileged factual statements, while others reveal legal theories or mental impressions. Redaction, segregation, metadata review, privilege logging, and in camera submission may be preferable to an all-or-nothing position.
A defensible analysis ultimately examines five related considerations: purpose, direction, content, confidentiality, and control. Purpose asks whether the material was created because of litigation. Direction asks whether counsel or another authorized representative requested or supervised the work. Content asks whether the record contains facts, strategic selections, mental impressions, legal theories, or some combination of them. Confidentiality asks whether the material was handled in a manner consistent with protecting it from adversaries. Control asks whether the firm understood and managed the provider’s access, retention, training, and disclosure practices.
No single consideration will decide every case. A prompt created by an attorney during active litigation may still lose protection if it is publicly posted. A client’s independently generated chat may remain protected if it was clearly created to prepare for litigation and kept confidential. An output that contains an ordinary factual chronology may receive less protection than a prompt sequence revealing counsel’s assessments and trial strategy. Courts will likely continue evaluating the entire factual setting rather than relying on the label attached to the technology.
Generative AI does not eliminate the attorney work-product doctrine, but it makes careful application of that doctrine more important. Prompts, outputs, uploaded files, and chat histories may contain protected litigation preparation, including counsel’s most sensitive mental impressions. They may also contain discoverable facts, unprotected business material, or communications whose confidentiality was compromised by the manner in which the technology was used.
The emerging decisions show that neither side of the categorical debate is persuasive. It is too broad to argue that everything submitted to or generated by AI is protected. It is equally broad to argue that use of a third-party AI platform automatically waives all work-product protection. Traditional principles remain capable of addressing the technology when courts examine the real purpose, direction, content, confidentiality, and control surrounding the use.
For attorneys and clients, the best protection will come from disciplined practices rather than aggressive labels. Secure platforms, documented litigation purposes, counsel-directed workflows, careful vendor review, meaningful supervision, protective-order provisions, preservation procedures, and accurate privilege logs can make AI-assisted legal work both useful and defensible. Generative AI may change how litigation materials are created, but the underlying interest protected by the work-product doctrine remains the same: allowing parties and their counsel to investigate, analyze, and prepare their cases without exposing their strategic thought processes to their adversaries.
This article is provided for general informational purposes and does not constitute legal advice.
Contact Tishkoff
Tishkoff PLC specializes in business law and litigation. For inquiries, contact us at www.tish.law/contact/. & check out Tishkoff PLC’s Website (www.Tish.Law/), eBooks (www.Tish.Law/e-books), Blogs (www.Tish.Law/blog) and References (www.Tish.Law/resources).
Footnoted Sources
1- Hickman v. Taylor, 329 U.S. 495 (1947). https://supreme.justia.com/cases/federal/us/329/495/
2- Federal Rules of Civil Procedure 26(b)(3) and 26(b)(5). https://www.law.cornell.edu/rules/frcp/rule_26
3- Federal Rule of Evidence 502. https://www.law.cornell.edu/rules/fre/rule_502
4- American Bar Association Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512, “Generative Artificial Intelligence Tools,” July 29, 2024. https://www.americanbar.org/groups/business_law/resources/business-law-today/2024-october/aba-ethics-opinion-generative-ai-offers-useful-framework/
5- State Bar of Michigan, “Artificial Intelligence for Attorneys—Frequently Asked Questions.” https://www.michbar.org/opinions/ethics/AIFAQs
6- United States v. Heppner, No. 25 Cr. 503 (JSR), 2026 BL 52143, 2026 U.S. Dist. EXIS 32697 (S.D.N.Y. Feb. 17, 2026). https://www.hrlegalist.com/wp-content/uploads/sites/4/2026/03/1-United-States-v.-Heppner.pdf
7- Warner v. Gilbarco, Inc., No. 2:24-cv-12333, 2026 WL 373043 (E.D. Mich. Feb. 10, 2026). https://law.justia.com/cases/federal/district-courts/michigan/miedce/2:2024cv12333/379552/94/
8- Morgan v. V2X, Inc., No. 25-cv-01991-SKC-MDB, 2026 WL 864223 (D. Colo. Mar. 30, 2026). https://www.everlaw.com/blog/ai-and-law/morgan-v-v2x-ai-disclosure-in-discovery/
9- Tate Group Automotive, LLC v. Legacy Automotive Capital, LLC, No. 25-BC11B-0020 (Tex. Bus. Ct. 11th Div. June 3, 2026). https://www.nelsonmullins.com/insights/blogs/corporate-governance-insights/all/everything-s-bigger-in-texas-including-work-product-protection-for-ai-chats
10- Assini v. Hayward, 2026 N.Y. Slip Op. 26086 (Sup. Ct. Nassau Cnty. June 4, 2026). https://www.nycourts.gov/reporter/current/3dseries/2026/2026_26086.shtml