From AI Hallucinations to Defensible Verification: A NYSBA CLE on New York Part 161
- Pouya Shafabakhsh
- Aug 24
- 5 min read
Artificial intelligence is already part of legal research, drafting, document review, and court-paper preparation. The more important question for litigators is increasingly not whether AI was used, but whether the work produced with its assistance can be independently verified before it reaches the court.
On September 24, 2026, from 12:00 to 1:00 p.m. ET, I will have the privilege of co-presenting a live New York State Bar Association CLE program with Hon. Ronald J. Hedges, former U.S. Magistrate Judge for the District of New Jersey:
“What Attorneys and Their Staff Might Do to Avoid AI Hallucinations: A Pre-Filing Verification Protocol Under Part 161.”
The program will focus on a practical problem facing attorneys and legal teams: how to move AI verification from an informal final check toward a structured, source-based pre-filing process.
Why New York Part 161 Matters
New York's Part 161, Use of Artificial Intelligence Technology, became effective June 1, 2026 and establishes a statewide policy applicable across the New York State Unified Court System in both civil and criminal matters.
The rule takes an important position: the use of AI tools in preparing court papers is not prohibited merely because AI is involved. Part 161 also states that attorneys and parties should not be required to disclose AI use simply because an AI tool assisted in preparing a submission.
But that should not be mistaken for reduced accountability.
Part 161 preserves the duties and responsibilities already attached to papers submitted to a court. It also permits individual courts, in their discretion, to adopt a part rule based on the Appendix A Model Rule.
Where that Model Rule applies, an attorney or party using AI in preparing a court paper must carefully review the submission and independently ensure that it contains no fabricated or fictitious cases, statutes, or other material. Signing the paper certifies that this review has been conducted.
That distinction matters.
Part 161 does not create a universal requirement that every New York court adopt Appendix A. Rather, it establishes the statewide framework while allowing courts to implement the Model Rule where appropriate.
For litigation teams, the operational question therefore becomes: What should a reliable AI-assisted pre-filing review actually look like?
AI Hallucinations Are Only Part of the Verification Problem
Much of the legal discussion about generative AI has focused on fabricated cases.
That is understandable, but a meaningful pre-filing review must go further.
An AI-assisted draft can contain a real case cited for a proposition the decision does not actually support. A quotation can be subtly inaccurate. A valid statute can be outdated or applied outside its relevant jurisdiction. Factual assertions can lose their connection to the underlying record. Secondary sources may appear authoritative while masking weaknesses in primary authority.
The official NYSBA program therefore identifies several areas participants will examine, including fabricated authority, inaccurate quotations and unsupported propositions.
The challenge is not simply asking:
“Did the AI hallucinate?”
It is asking:
“Can every consequential proposition in this filing be independently traced back to a reliable source?”
That is a much stronger verification standard.
A Source-Based Six-Stage Verification Process
During the CLE, we will examine a source-based six-stage verification sequence for AI-assisted court papers.
The purpose is not to add unnecessary bureaucracy to litigation practice. It is to create a repeatable pathway between an AI-assisted draft and the attorney who ultimately signs the submission.
The verification process is designed around a simple principle: important legal propositions should be tested against their authoritative sources rather than accepted because a model produced text that appears legally persuasive.
Participants will examine how a structured workflow can help legal teams identify where verification is required, distinguish primary from secondary sources, test quotations and propositions against original authority, review factual assertions, and document the allocation of review responsibilities.
NYSBA's published learning objectives specifically identify the ability to apply a source-based six-stage verification sequence as a program outcome.
Delegation Does Not Mean Transfer of Responsibility
Modern litigation is collaborative.
Associates, law clerks, paralegals, litigation-support teams, researchers and technology professionals may all participate in preparing a filing. AI introduces another layer into that workflow.
That makes allocation of review tasks important—but it also makes the distinction between delegation and responsibility critical.
A law firm may divide verification work among team members. Different individuals may check citations, quotations, factual assertions or source materials.
But the workflow should not obscure who ultimately takes responsibility for the court submission.
One of the central learning objectives of the September 24 program is therefore to examine how firms can allocate and document review tasks while preserving the signing attorney's independent responsibility.
That is where documentation becomes more than administration.
A structured verification record can help establish what was reviewed, which sources were consulted, who performed particular checks and whether identified problems were resolved before filing.
For firms supervising increasing amounts of AI-assisted legal work, that type of reconstruction may become increasingly important.
From Informal Checking to a Defensible Pre-Filing Process
Many lawyers already verify AI-assisted work.
The more difficult question is whether the process is sufficiently structured to be explained later.
Imagine being asked several months after a filing:
Which propositions were independently checked?
Against which sources?
Who reviewed the citations?
Were quotations compared with the original decisions?
What happened when an inconsistency was found?
Who performed the final review?
A process that exists primarily in the memories of individual team members may be difficult to reconstruct.
A more defensible approach treats verification as a defined professional workflow rather than an assumption that somebody checked the output before filing.
This is the practical territory we intend to explore during the CLE.
Who Should Attend?
The program is particularly relevant for:
New York litigators using generative AI in research or drafting;
attorneys supervising associates or staff who use AI tools;
managing partners and practice-group leaders establishing internal AI procedures;
litigation-support and legal-operations professionals involved in AI-assisted workflows; and
lawyers seeking a practical understanding of how Part 161 and the Appendix A Model Rule affect court-paper preparation.
The objective is not to argue that attorneys should avoid AI.
It is to examine how AI-assisted legal work can remain consistent with the professional responsibility that accompanies submission of a paper to a court.
NYSBA CLE Program Details
Program: What Attorneys and Their Staff Might Do to Avoid AI Hallucinations: A Pre-Filing Verification Protocol Under Part 161
Date: September 24, 2026
Time: 12:00–1:00 p.m. ET
Format: Live virtual webinar
CLE Credit: 1.0 Ethics and Professionalism credit
Faculty:
Hon. Ronald J. Hedges — Former U.S. Magistrate Judge, U.S. District Court for the District of New Jersey
Pouya Shafabakhsh — Principal Judicial Forensic AI Auditor | Chief AI Governance Officer, CAIC, AIGP
The program is sponsored through NYSBA's Committee on Continuing Legal Education, Committee on Artificial Intelligence and Emerging Technologies, and Committee on Law Practice Management. NYSBA currently lists the program as free for members.
I extend my sincere gratitude to Hon. Ronald J. Hedges for the opportunity to co-present this program.
I look forward to exchanging perspectives with fellow NYSBA members, New York litigators, and attorneys who are already confronting the practical reality of AI-assisted legal work.
The question is no longer simply whether AI can help prepare a court paper.
The more consequential question is:
Before the attorney signs it, can the team independently demonstrate that what matters has been verified?
We appreciate the completion of the Assessment Form at:
Author: Pouya Shafabakhsh Co-Founder, CAIO & Principal Forensic AI Auditor, Radsam Academy of AI Sovereign Governance. The Architect of North America's: Judicial Forensic AI Audit Standards, AI Governance, Risks & Compliance Standards, Air-Gapped Sovereign Sanctuary AI Audit System.
Comments