Seven AI Checks Before Serving a New York Expert Report
An expert may write every sentence and still rely on a record that artificial intelligence helped select.
By Pouya Shafabakhsh
Reprinted with permission from the September 29, 2026 edition of the “New York Law Journal” © 2026 ALM Global Properties, LLC. All rights reserved. Further duplication without permission is prohibited, contact 877-256-2472 or asset-and-logo-licensing@alm.com.
Consider a lost-profits dispute in New York’s Commercial Division. The damages expert writes the report without asking a chatbot to draft a paragraph. An assistant, however, has used generative AI to select the contracts, correspondence and spreadsheets the expert will examine. Asked whether AI wrote the report, the expert can truthfully say no. That answer leaves a part of the work unexplained.
This hypothetical illustrates a gap in checking only who wrote the report. AI may influence the analysis before drafting, when a tool determines which documents receive attention. A report can contain accurate quotations and correct calculations while resting on an incomplete selection of evidence.
For litigation leaders, the practical question is whether the expert can account for the route from the available records to the opinion.
New York has seen the problem of unexplained AI assistance. In Matter of Weber, a 2024 Saratoga County Surrogate’s Court decision, an expert used Microsoft Copilot to cross-check damages calculations but could not recall his prompts or identify the sources it used. The court had independently found his analysis unreliable before addressing AI. Its discussion nevertheless exposes a vulnerability: an expert who cannot explain a tool’s contribution leaves counsel with an avoidable gap.
For New York litigation teams, that gap deserves attention at retention, while there is time to examine the work. These seven checks offer guidance.
Identify where AI entered the work
“Did you use AI?” invites an answer too broad to be helpful. Editing, translation, document selection, coding and substantive analysis involve different kinds of dependence. An account of the workflow, including assistants’ contributions, can establish where a tool influenced the evidence or the conclusions. The inquiry concerns material involvement, not a census of every software feature on the expert’s computer.
Check what the expert never saw
Reviewing retrieved documents tests what the system found. It does not establish what the system missed. Where AI filtered a collection, the useful questions concern the selection criteria, the excluded material and any independent checks for omissions. A check might examine whether the process retrieved records supporting a competing explanation. Its scope will depend on the assignment and the collection.
Review the task behind the prompt
A request to identify evidence supporting a hypothesis differs from an instruction to test it against alternatives. Neither prompt, viewed alone, establishes whether the opinion is sound. An expert may investigate one possibility before rejecting it. The sequence matters: what was tested next, what contradicted the initial position and whether the analysis changed. Reviewing isolated language risks mistaking an intermediate exercise for a conclusion.
Verify the work independently
Asking the same tool whether its answer is accurate adds little assurance by itself. Independent verification connects a quotation to its source, a calculation to its inputs and a claim to the expert’s methods. Review depth can follow the significance of the output. An error in punctuation and an error in the assumptions supporting a damages estimate warrant different attention.
Keep the method explainable
An expert need not explain a model’s internal architecture to describe how it was used. The account identifies the task, inputs, limitations, checks and reasons for accepting or rejecting results. In New York’s federal courts, Rule 702 addresses sufficient facts or data, reliable principles and methods, and their reliable application. The governance objective is to make the expert’s contribution understandable, without confusing technical documentation with an admissibility determination.
Locate the underlying records
A conversation history, uploaded file, generated spreadsheet and final report may be stored in different places. Identifying those locations early gives counsel a basis for assessing preservation and potential production. Missing information should remain identified as missing. A later reconstruction can be useful if clearly described, but it cannot honestly be presented as a record made during the original process.
Match disclosure to the forum
Disclosure requirements vary with the forum. CPLR 3101(d)(1)(i) provides for disclosure, on request, of specified information about an expert’s expected testimony, including the facts, opinions and grounds. In the Commercial Division, Rule 13(c) generally requires signed reports from retained experts, including the data or other information considered, unless otherwise stipulated or ordered. In federal litigation, Rule 26 requires specified expert reports while protecting drafts and certain attorney-expert communications, subject to exceptions. Counsel must assess AI records within the applicable framework and court orders. The word “prompt” supplies no automatic answer about discoverability.
These checks are governance considerations, not a proposal to demand every keystroke or treat AI use as evidence of unreliable work. Well-designed assistance may help an expert organize a record, identify inconsistencies or examine an alternative thoroughly. The review should establish which results the expert checked and which limitations still affect the opinion.
For a New York litigation chair, the practical change is to assign responsibility for this conversation when the expert is retained and revisit it before disclosure. If AI selected documents, someone should examine how omissions were tested. If it supplied calculations, the expert should explain the independent checks. Unresolved answers need time in the schedule, before the report reaches opposing counsel.
Counsel can put the seven checks on the agenda for the first expert meeting. Before service, return to any unanswered question, identify the person responsible and confirm whether the report needs correction. That makes review part of preparing the evidence, with time to address weaknesses before they become a dispute.
Pouya Shafabakhsh is a Principal Forensic AI Auditor and co-founder of Radsam Academy of AI Sovereign Governance. He is not a lawyer. This article provides independent analysis and opinion, not legal advice.
Reprinted with permission from the September 29, 2026 edition of the “New York Law Journal” © 2026 ALM Global Properties, LLC. All rights reserved. Further duplication without permission is prohibited, contact 877-256-2472 or asset-and-logo-licensing@alm.com.
Author: Pouya Shafabakhsh Co-Founder, CAIO & Principal Forensic AI Auditor, Radsam Academy of AI Sovereign Governance. Independent AI governance and forensic-audit analysis. Educational information only; not legal advice.
Would you be opposed to finding your firm's AI gaps before opposing counsel or insurance companies do? I invite you to assess your firm using our AI Risks Calculator. Safely, securely, privately, and confidentially—with no need to enter any information, not even your email address—you can get a comprehensive AI Risks and Liabilities report in just five minutes.
If you have a sensitive file and want to protect it against AI hallucinations, bias, algorithmic discrimination, and AI-assisted risks through an Air-Gapped Shadow AI Audit before court, or if you are involved in a legal dispute related to AI and need a Judicial Forensic AI Audit, at Radsam, we are willing to assist you whether through our human-based services or by providing our in-house solution, the Sovereign Sanctuary Vault. In these cases, I invite you to fill out our Pre-Qualifying Assessment Form. We will review your information, qualifications, and capacity, and will contact you within two business days.
Comments