NYCLA Op. 755-2026 – Basic Ethical Obligations All Lawyers and Recommendations for Solo and Small-firm Practitioners Concerning the Use of Generative Artificial Intelligence

NEW YORK COUNTY LAWYERS’ ASSOCIATION

COMMITTEE ON PROFESSIONAL ETHICS

Opinion No.:  755- 2026

Date: November 5, 2026

Re: Basic Ethical Obligations All Lawyers and Recommendations for Solo and Small-firm Practitioners Concerning the Use of Generative Artificial Intelligence

 

QUESTION PRESENTED

What are the basic ethical obligations concerning the use of generative artificial intelligence in litigation, document preparation, and court submissions? What implications exist for attorney-client privilege, work-product waiver, confidentiality, and competence under the New York Rules of Professional Conduct (“NY RPC”), as well as under 22 NYCRR Part 130 and Part 161 of the Rules of the Chief Administrator to the extent those regulations overlap with the NY RPC? How can solo and small-firm practitioners utilize the benefits of generative artificial intelligence in a manner compliant with their ethical obligations?

OPINION

 The rapid integration of generative artificial intelligence (GenAI) and Large Language Models (LLMs) into legal practice presents distinct ethical challenges for all practitioners, but the operational risks are perhaps even greater for solo practitioners and small firms. While the Unified Court System explicitly permits the use of AI tools in document preparation, attorneys maintain a non-delegable, personal duty to ensure that the use of these tools does not compromise attorney-client privilege or work-product protections, violate client confidentiality, or result in frivolous or unsupported court submissions.

Under New York Rules of Professional Conduct (NY RPC) 1.1 and 1.6, an attorney must perform a rigorous pre-deployment assessment of any AI tool to ensure its security protocols protect against the waiver of privilege and work-product protection. Inputting sensitive or unredacted client data into an unsecured, public AI tool constitutes a voluntary disclosure to a third party, destroying the expectation of confidentiality necessary to maintain legal privileges. Furthermore, under NY RPC 3.3 and 22 NYCRR Part 130, signing and submitting a court paper containing AI-generated “hallucinations” constitutes “frivolous” conduct.

To satisfy these intersecting mandates, solo and small-firm practitioners face specific challenges in implementing sufficient safeguards to ensure compliance with the NY RPCs. While solo and small-firm practitioners face increased risk due to lack of staff support or fewer layers of review, all lawyers regardless of their organization’s size or nature should be cognizant of the possible ethics violations which arise when AI software is employed in the practice of law. This opinion describes the minimal steps which can ensure compliance, beginning with a strict verification of the AI software prior to tool selection, followed by manual independent verification of all AI-assisted work product before signature and filing. Far from seeking to dissuade attorneys from using AI tools, this opinion seeks to both bring to light the possible ethical issues which can arise from their use and ensure that attorneys are well prepared to safely implement AI within their practice. To this end, the elements in this checklist ensure that no disclosure of privileged information occurs, and that attorneys adhere to the NY RPC as they use AI software. Of course, the nature of technological development—especially that surrounding GenAI—means that this opinion is not intended to, or will be, the “last word” on the matter. Bar associations have published guidance on the ethical use of GenAI for either general purposes (e.g., Am. Bar Ass’n, Formal Op. 512: Generative Artificial Intelligence Tools) or specific use cases (e.g., N.Y.C. Bar Ass’n Formal Op. 2026-2: Ethical Use of AI for Recording, Transcribing, and Summarizing Non-Client Conversations). This opinion is therefore meant as a blueprint, reflecting a snapshot in time, to empower practitioners to feel confident in embracing this new technology in a manner consistent with their obligations under the NY RPC.

I.  RECENT LEGAL DEVELOPMENTS

 The New York legal landscape is experiencing a significant technological transformation driven by the proliferation of GenAI tools capable of drafting pleadings, synthesizing voluminous discovery records, and generating legal memoranda, among other tasks. As recognized by New York courts, this technological evolution represents a new paradigm for the legal profession—one which is not inherently improper, but rather has the potential to offer substantial benefits to attorneys, clients, and the broader public by increasing administrative efficiency and reducing delivery costs (Deutsche Bank Natl. Trust Co. v. LeTennier, 251 A.D. 3d 97 [3d Dept 2026]).

Recognizing this reality, the Unified Court System (UCS) enacted a set of preliminary rules. Pursuant to the Administrative Order of the Chief Administrative Judge of the Courts (AO-75-26), a new Part 161 of the Rules of the Chief Administrator (22 NYCRR §§ 161.1 to 161.4 and Appendix A) became effective on June 1, 2026. Part 161 codifies the statewide policy that the use of artificial intelligence tools by attorneys in preparing court papers should not be prohibited (§ 161.3). Crucially, the rule establishes that because standard ethical duties already apply to all submissions, attorneys are not required, as a matter of default statewide procedure, to disclose to the court that they have specifically utilized AI in the preparation of their papers (id.).

However, Part 161 explicitly preserves the discretion of individual judges to adopt localized rules—such as the model certification rule set forth in Appendix A—requiring explicit AI disclosure and verification certificates. More fundamentally, Appendix A reminds practitioners that under existing signature and certification authority (e.g., 22 NYCRR 130-1.1, 130-1.1a), an attorney who signs and files any document with a New York court affirmatively certifies that the paper contains no false material factual statements, frivolous legal arguments, or fabricated content.

Parallel to these state court developments, federal courts have increasingly confronted the severe evidentiary consequences of misusing GenAI tools. In early 2026, the United States District Court for the Southern District of New York issued a precedent-setting ruling addressing the privilege and work-product implications of using unsecured, public AI tools (see United States v. Heppner, 820 F. Supp. 3d 292 [SD NY 2026]). In that case, the court ruled that documents prepared by a public AI platform were not protected by the attorney-client or work product privilege where the platform’s terms of service permitted data sharing with third parties. Moreover, individual federal judges have taken the initiative to amend their Individual Practices and require attorneys to disclose their use of GenAI for filings in cases over which they preside. Therefore, attorneys should always check the latest version of a federal judge’s Individual Practices to ensure compliance with any particular rules governing the use of GenAI.

This Committee issues this opinion to address the acute, structural risks that this new regulatory and evidentiary environment poses to solo practitioners and small law firms. While technology can act as a significant economic equalizer, a lack of institutional safeguards in boutique or small practices can rapidly transform a standard drafting oversight into a devastating waiver of privilege, an administrative disciplinary violation, or a ground for imposing personal sanctions. As technology evolves, so must attorneys—in order to remain in compliance with the NY RPC, attorneys must learn to navigate an increasingly complex technological landscape. This opinion does not seek to impose restrictions which bind attorneys more tightly than New York courts have been willing to thus far—it simply aims to offer structured guidance for attorneys striving to navigate ethically this rapidly evolving technological landscape.

 

II.  SOLO/SMALL-FIRM PRACTICE VS. LARGE FIRM INFRASTRUCTURE

 The ethical demands of these new rules and recent federal evidentiary case law are disproportionately harsh on the solo and small-firm practitioner. Large corporate law firms possess substantial institutional infrastructure designed to absorb and mitigate technological risks. These entities routinely execute enterprise-level licensing agreements with specialized legal-AI vendors (such as Harvey or CoCounsel). These high-cost, bespoke platforms include ironclad data privacy addenda preventing vendor ingestion of client data, employ proprietary hallucination-filtering mechanisms, and frequently feature corporate vendor-backed indemnification policies. Moreover, automated work product within a large firm is insulated by multi-layered human review hierarchies, passing from junior associates to senior associates, dedicated citation-checkers, supervising partners and, ultimately, managing partners, all of whom can consult with their firm’s inhouse counsel. Melia Robinson, Inside the AI Boom That’s Changing How Big Law Attorneys Work, Business Insider, July 18, 2025, https://www.businessinsider.com/big-law-top-10-firms-ai-overhaul-use-cases-2025-7.

In stark contrast, solo practitioners and small-firm attorneys operate without dedicated IT departments, internal general counsel, or multi-tiered associate review structures. They function as a “team of one,” personally managing research, drafting, calendaring, and final proofreading. To maintain low overhead and offer competitive rates, small-firm practitioners frequently rely on consumer-grade or retail-level AI software (such as public subscriptions to commercial LLMs). While this software can make research and drafting more efficient, attorneys should thoroughly verify the technology they use and the outputs it produces, maintain professional judgment when using AI software, and avoid over-reliance on these tools.

Consumer-facing platforms operate under terms of service that differ drastically from legal enterprise contracts. By default, they retain user prompt histories to train public models, and they lack localized legal citation verification networks. Consequently, when a resource-constrained solo practitioner utilizes a public AI tool to meet a compressed motion deadline, there is little or no institutional safety net to catch an operational oversight.  The unverified technological error or un-redacted disclosure passes directly into the public sphere or final court submission, often imposing severe professional liability and personal financial exposure directly onto the individual practitioner.

However, while solo and small-firm practitioners may need to perform more up-front work in verifying the appropriateness of customer-facing AI tools than large firms might in using proprietary software, it is imperative that all attorneys adhere to the recommendations set forth in this opinion in order to avoid running afoul of ethics rules. All legal practitioners face court-imposed time constraints and client demands for cost-efficiency. Therefore, regardless of whether the AI platform used is proprietary or consumer-grade, its outputs must be independently verified in order to preserve the attorney’s professional judgment.

It is also worthwhile to note that, while attorneys in large firms may have access to specialized IT and cybersecurity teams in choosing and operating AI software, small-firm and solo practitioners are often self-reliant in that regard. Because ethical compliance cannot be conditioned on resources which an attorney may or may not have access to, this opinion provides a checklist which can be implemented regardless of firm size or budgetary constraints.

 

A.  NY RPC 1.1 (Competence) and the Duty of Technological Literacy

NY RPC 1.1(a) mandates that a lawyer shall provide competent representation to a client, which requires the legal knowledge, skill, thoroughness, and preparation reasonably necessary for the representation. Comment [8] to NY RPC 1.1 explicitly extends this obligation to the technological arena, stating that to maintain the requisite knowledge and skill, a lawyer should keep abreast of the benefits and risks associated with relevant technology.

Competence in the modern era does not demand that a small-firm practitioner possess an advanced degree in computer science. It does, however, require a foundational understanding of the mechanical and functional limitations of the specific software tools integrated into their practice. GenAI models are LLMs built upon probabilistic text-prediction architecture. They are designed to predict and compile the most statistically probable sequence of words responding to a user’s prompt, based on their underlying training data corpus. They are not deterministic relational databases, such as traditional Westlaw, LexisNexis, or official state case-law reporters. Nor do they engage in complex reasoning and analysis, as humans do. Fundamentally, LLMs, by their very nature, are backward-looking. This is because an LLM “treats pretrained data as constituting the known hypothesis space.” Amy Wenxuan Ding & Shibo Li, Generative AI Lacks the Human Creativity to Achieve Scientific Discovery from Scratch, Scientific Reports, Mar. 20, 2025, https://www.nature.com/articles/s41598-025-93794-9. Therefore, an LLM is unable to generate content that would require breaching the boundaries of its underlying training data corpus: in other words, it is only as good as the data it has access to.

Because LLMs prioritize linguistic coherence and syntactic persuasion over historical fact, they are structurally prone to generating “hallucinations”—the fabrication of authoritative-sounding but entirely fictitious case names, docket numbers, volume citations, and verbatim judicial quotations. Presenting an LLM with a prompt such as “Find me a Second Department case supporting a motion to dismiss a constructive trust claim” will often result in a perfectly formatted, highly persuasive, but completely non-existent judicial opinion.

This Committee emphasizes that treating an unverified GenAI tool as a primary legal research mechanism or an automated citation service constitutes a severe breach of the duty of competence under NY RPC 1.1. Competence dictates that a solo practitioner must treat any AI-generated text or citation with the same extreme skepticism they would afford to an unverified draft prepared by an untrained, novice clerk. Relying blindly on technology without manual, independent human verification constitutes a per se failure of adequate preparation and thoroughness under NY RPC 1.1(a) (see also Park v. Kim, 91 F.4th 610 [2d Cir 2024] [referring counsel to a disciplinary committee for relying on an unverified, AI-hallucinated case citation in appellate briefing]).

In order to fulfill this duty of competence, attorneys should not neglect their professional judgment when working with AI software. Cases and citations should be independently vetted before being included in filings. Not only must an attorney verify that the cases cited by the AI tool exist, but also that the facts are relayed accurately by the AI software. An attorney must assess whether the AI software’s account of a case’s holding actually tracks that of the court in question, or whether the software has utilized logical leaps—that may indeed be sound—in its presenting of a holding. Attorneys should also ensure that the cases cited by the AI platform are still good law. Similarly, attorneys should keep in mind that AI software cannot make strategic decisions which overtake the attorney’s professional judgment. It remains the attorney’s duty to establish legal strategies, to provide accurate information to clients and courts, and to ensure that filings rely on existing and relevant authorities. It is very possible that the AI software may suggest an argument, by analogy or logical extrapolation, that an attorney would want to include in a filing. However, the attorney must be the one to determine whether it passes the “sniff test” in light of existing case law, academic scholarship, and a particular court’s own precedents.

B.  NY RPC 1.6 (Confidentiality)

The protection of attorney-client confidentiality and attorney work product protection is central to our litigation process. Critical to both the attorney-client privilege and the work product doctrine is the maintenance of confidentiality within the confines of NYRPC 1.6, the CPLR and applicable case law. Typically, disclosures of privileged information to third parties – and especially adverse third-parties — will negate both protections.

For the solo and small-firm practitioner, this represents an existential hazard. A single late-night upload of a client’s narrative summary into an unsecured platform can completely strip that client of their confidentiality and work product protections. The lawyer’s ethical obligation under NY RPC 1.6 (alongside the NY RPC 1.1 duty of competence already discussed) requires a rigorous, preemptive vetting of the tool’s data-security architecture before any text is entered. For example, attorneys should remove identifying client information from prompts given to AI software, redact any privileged or case-specific information from prompts, and omit information which may affect negotiations or the terms of a settlement.

NY RPC 1.6(a) states that a lawyer shall not knowingly reveal confidential information or use such information to the disadvantage of a client or for the advantage of the lawyer or a third person, unless the client gives informed consent. Confidential information encompasses any information gained during or relating to the representation of a client, whatever its source, that is protected by the attorney-client privilege, is likely to be embarrassing or detrimental to the client if disclosed, or information that the client has requested be kept confidential.

The use of consumer-grade, retail AI tools creates an immediate and severe structural hazard for client confidentiality under NY RPC 1.6. The standard commercial terms of service for public AI platforms explicitly state that any data entered into the prompt field is captured, retained, and analyzed by the developer’s engineers and algorithms to train future iterations of the public model.

When a solo practitioner types a highly detailed, proprietary fact pattern into a public AI prompt box to “draft an opposition to a motion for summary judgment,” or uploads an unredacted deposition transcript to “generate a multi-page summary,” the attorney has actively published confidential client data to a third-party corporate entity. Because this disclosure is made without a confidentiality or non-disclosure agreement that binds the vendor, and without the client’s explicit, informed consent, it represents a direct violation of NY RPC 1.6(a).

The fact that the disclosure was motivated by a desire to reduce billable hours or increase drafting speed for the client’s benefit is entirely immaterial. Small-firm practitioners must recognize that unless they are operating within an interface where data-training settings are explicitly deactivated (e.g., “Do Not Train” toggles or dedicated legal enterprise platforms), they must completely sanitize all inputs of any identifying information, client names, specific geographic markers, or unique transactional characteristics before interacting with the tool.

C.  NY RPC 3.3 (Candor Before the Tribunal), Part 130 Certifications, and Emerging Precedent

NY RPC 3.3(a)(1) dictates that a lawyer shall not knowingly make a false statement of fact or law to a tribunal or fail to correct a material false statement of fact or law previously made to the tribunal by the lawyer. While “knowledge” under the rules requires actual awareness of the fact, NY RPC 1.0(k) clarifies that a person’s knowledge may be inferred from circumstantial evidence.

Under New York practice, the ethical obligations of NY RPC 3.3 are directly intertwined with the regulatory mandates of 22 NYCRR Part 130. Pursuant to § 130-1.1a, every court paper filed or submitted must be signed by an attorney of record. This signature acts as an affirmative certification by the attorney that, to the best of that person’s knowledge, information, and belief, formed after an inquiry reasonable under the circumstances: (1) the presentation of the paper or the contentions therein are not frivolous; and (2) the statements of fact contained therein are true. Frivolous conduct includes the assertion of legal arguments that are completely without merit in law and cannot be supported by a reasonable argument for an extension, modification, or reversal of existing law, or conduct that involves the assertion of material statements of fact that are false (§ 130-1.1[c]).

The adoption of 22 NYCRR Part 161 (Appendix A) explicitly maps these historical rules onto GenAI usage. Appendix A mandates that any attorney utilizing an AI tool in preparing any paper filed in or submitted to the court is required to carefully review the paper and independently verify the accuracy and reliability of any statement or citation. The rule concludes with a definitive standard: “By signing such paper, an attorney or party certifies that such a review has been conducted and that the paper contains no such fabricated or fictitious content.”

The New York judiciary enforces these obligations strictly, demonstrating that a failure to manually cross-check AI output crosses the threshold from simple negligence into sanctionable, reckless misconduct. In the landmark decision Deutsche Bank Natl. Trust Co. v. LeTennier, 251 A.D. 3d 97 [3d Dept 2026], the Appellate Division issued what stands as the state’s first appellate-level award of sanctions for improper AI usage. In LeTennier, a defense attorney submitted an appellate brief in a routine foreclosure action that contained six entirely nonexistent case citations and multiple false legal propositions drawn from AI hallucinations.

Crucially, when the plaintiff’s counsel identified the fictions and moved for sanctions, the defense attorney failed to perform basic, manual due diligence. Instead, the attorney filed a reply brief and subsequent letters to the court that contained additional fabricated cases, alongside gross misrepresentations regarding the holdings of real, existing judicial opinions. The Third Department held that the unverified usage of GenAI to draft appellate briefs containing false and hallucinated information constitutes frivolous conduct under 22 NYCRR 130-1.1, warranting the imposition of financial sanctions. The court flatly rejected the notion that technological errors mitigate an attorney’s exposure, emphasizing that a lawyer’s signature certifies personal, independent verification.

This enforcement posture was further solidified by the Appellate Division, Second Department, in Landberg v. City of New York, 262 N.Y.S. 3d 543 [2d Dept 2026]. There, the court ordered personal monetary sanctions totaling $10,500—imposing an $8,000 fine on the individual attorney and a $2,500 fine on his law firm, payable to the Lawyers’ Fund for Client Protection. The court found that the attorney had submitted a civil appellate brief prepared with the assistance of GenAI containing multiple citations to non-existent cases, fictitious Court of Appeals quotations that were wholly contrary to actual law, and severe misrepresentations of real holdings.

These cases establish a strong expectation for the small-firm bar: when an attorney signs a document under 22 NYCRR § 130-1.1 containing AI-generated fabrications, the failure to perform the foundational task of looking up the cases in a verified legal database provides the necessary circumstantial evidence to infer reckless disregard. This recklessness satisfies the constructive knowledge threshold under NY RPC 3.3 and transforms the submission into a sanctionable, frivolous filing under Part 130. As exemplified here, attorneys in firms of all sizes are vulnerable to disciplinary action and sanctions linked to AI misuse. While a large firm may be able to absorb penalties or reprimands as operational setbacks, solo and small-firm practitioners may face these as severe and potentially career-ending events. This disparity underscores the necessity of critically evaluating AI software outputs in order to ensure no misrepresentations are made or frivolous charges brought before a tribunal.

It must be remembered that while the technology is new, the concept underlying the ethics rules is anything but: when an attorney puts their signature to a document, they are certifying the accuracy of its contents. When drafting anything, there is no difference between an attorney that uses pen-and-paper and an attorney that uses AI software when the documents are ultimately certified by their respective signatures. As many practitioners will admonish their clients, signing something without verifying its truthfulness and accuracy has always been a dangerous game to play.

 

D.  Non-Delegable Accountability and the Supervision of Assistants

Solo practitioners and partners in small firms frequently utilize temporary law clerks, paralegals, or independent contract attorneys to manage excess drafting burdens. This operational model introduces secondary ethical exposure under NY RPC 5.1 and 5.3. NY RPC 5.1 requires partners and supervisory lawyers to make reasonable efforts to ensure that subordinate lawyers conform to the Rules of Professional Conduct, while NY RPC 5.3 extends this non-delegable supervisory mandate to non-lawyer assistants.

If a solo practitioner delegates a motion brief to a freelance paralegal or an independent contract attorney, and that assistant utilizes an unverified AI model to complete the task, the principal attorney remains entirely exposed. As illustrated in the federal enforcement decision Matt McCormick v. Texakoma Fin., Inc., 839 F. Supp. 3d 549 [ED Tex 2026], courts will flatly reject attempts by a signing attorney to deflect blame onto departing clerks, paralegals, or independent contractors. In McCormick, despite counsel’s plea that she had practiced for over twenty years without incident and that her clerk had committed the AI drafting error, the court issued a public reprimand, ordered publication in the Federal Supplement, and required the attorney to personally submit a signed “Certification of Verification” in every filing for the remainder of the calendar year.

The lesson for the small-firm bar is absolute: under NY RPC 5.1 and 5.3, supervisory accountability is non-delegable. An attorney’s signature on a pleading or motion is a personal guarantee of accuracy that cannot be insulated by blaming a subordinate’s reliance on technology. Additionally, an AI technology’s output cannot be treated as authoritative guidance on legal matters and should similarly be scrutinized and reviewed by a supervising attorney.

Again, this concept should be nothing new to any legal practitioner. Paralegals, law clerks, and law school interns provide crucial services in the running of practices regardless of organizational structure: from the solo family-run law office to the federal government. No matter the surroundings, baked into NY RPC 5.1 and 5.3 is the concept that inexperienced attorneys and non-members of the Bar can only provide so much assistance and, ultimately, it is the supervising attorney on any project that shoulders the responsibility of ensuring that filings are accurate and proper.

 

III.  OPERATIONAL COMPLIANCE GUIDANCE

To ensure that solo and small-firm practitioners can harness the efficiency of GenAI without exposing their clients to privilege waivers or themselves to personal sanctions, this Committee outlines a four-pillared, scale-appropriate compliance protocol that satisfies NY RPC 1.1, 1.6, 3.3, 5.1 and 5.3, while remaining consistent with Part 161.                  

Step 1: Pre-Deployment Privilege Verifications

Prior to entering any client narrative, case theory, or documentary evidence into an AI interface, a practitioner should systematically execute the following Privilege Verification Checklist derived from standard legal technology compliance guidelines (see New York State Academy of Trial Lawyers CLE, “Trust, But Verify: Ethical AI Use in Legal Practice,” June 23, 2026). Attorneys should verify these elements within the software’s terms of service and privacy policy in order to ensure compliance.

• [ ] Terms of Service (TOS) Audit: Review the contract or terms governing the tool to verify that the vendor explicitly guarantees a confidential relationship and assumes the legal status of an agent of the attorney, maintaining confidentiality matching NY RPC 1.6.

• [ ] Data Utilization & Training Opt-Out: Verify that the platform has an absolute, verified configuration ensuring that inputs and prompt histories are never utilized for public model training, algorithmic reinforcement, or vendor review. If an account does not permit a complete opt-out of data ingestion, it is classified as an unsecured public tool and cannot be used for confidential client operations.

• [ ] Data Retention & Purge Window: Identify the vendor’s data retention policy. Confirm whether data is deleted immediately upon session termination or stored on external servers. If stored, confirm that it is encrypted both in transit and at rest using legal-grade encryption standards.

• [ ] Sub-processor Disclosures: Verify whether the software app utilizes API calls to third-party providers (e.g., passing data from a localized application to an external developer’s servers). Ensure that identical confidentiality and non-ingestion agreements bind all underlying sub-processors.

• [ ] Written Client Disclosure/Informed Consent: If the tool fails to satisfy any of the security thresholds above, but the attorney still wishes to use the tool for computational assistance, the attorney must obtain the client’s explicit, informed consent in writing after disclosing the heightened risk of third-party privilege waiver under NY RPC 1.4.

Step 2: Manual Checking Procedures

No citation, quotation, or legal proposition generated with the assistance of an AI tool may be filed with a New York court or served upon an adversary unless it has been subjected to a manual, independent human cross-check by the signing attorney within an official, traditional legal database (e.g., Westlaw, LexisNexis, Fastcase, or the official New York State Law Reporting Bureau series). This filter requires separate verification of three distinct components:

• Verification of Existence: The practitioner must physically open the case file in a verified database to confirm that the case name, volume number, reporter, and docket year actually exist in the physical or digital historical record.

• Verification of Quotes: Any text enclosed in quotation marks that is generated by an AI model must be checked word-for-word against the text of the official reporter. GenAI tools frequently synthesize distinct portions of an opinion into a single, fabricated sentence, presenting it as a direct quote.

• Verification of Substance and Holding: The practitioner must read the text of the verified opinion to ensure it stands for the precise legal proposition asserted in the brief. As demonstrated in LeTennier and Landberg, attorneys will face severe sanctions not only for citing non-existent cases, but for misstating the core holdings of real, existing judicial authorities.

• Verification of Opinion Status: The practitioner must ensure that any case cited by an AI model is current and has not been overturned or otherwise displaced at the time of citation.

Step 3: Complying with Data Privacy Requirements

 Before any client-related data is entered into an artificial intelligence interface, a solo or small-firm practitioner must execute one of the following protective measures:

• Confirm that the platform’s active configuration is governed by an enterprise agreement or a data privacy setting that explicitly deactivates conversational history storage and prohibits model training on user inputs.

• If operating within a standard, public retail interface where absolute data segregation cannot be guaranteed, the practitioner must systematically purge the input text of all confidential parameters. All client names, specific corporate entities, unique dates, local addresses, negotiation positions, and distinguishing financial or familial characteristics must be replaced with generic placeholders (e.g., “Client A,” “Company X,” “Date Y”).

Step 4: Applying a Formal Supervisory Policy

Solo practitioners and small law firm partners who utilize independent contract lawyers, temporary law clerks, or non-lawyer administrative assistants must establish and distribute a clear, written operational policy. This policy must explicitly forbid subordinates from submitting any AI-assisted draft to the principal attorney without attached, primary-source PDF copies of every cited authority or a written log confirming that the elements contained in this checklist were executed. The principal attorney must personally perform random or systematic audits of these verification logs before applying their physical or electronic signature to any final court submission under 22 NYCRR § 130-1.1.

Further, attorneys may want to preserve a record of AI usage within the firm, ensuring that they know on which matters the tools were used. It may also be useful to maintain a log of any incidents involving AI software in order to reevaluate the firm’s AI policy as needed. For example, attorneys may want to reevaluate their AI policy as soon as an incident arises in order to determine whether the policy has blind spots which can be fixed. In the absence of any incidents, a yearly review of the AI policy can ensure that any technological advances are taken into account.

Attorneys using consumer-grade AI should also be mindful of changes to the terms of service and privacy policy of the platform they employ. Regular verification of these terms will ensure that no inadvertent disclosure is made in the event of a change and will allow attorneys to proactively select an adequate alternative as needed. Managing attorneys should also enact policies whereby all staff who work under their supervision are aware of the risks associated with consumer-grade AI software updates. Solo practitioners and small law firm partners may want to undertake assessing each software update for compliance with the NY RPC. Therefore, it is crucial that all other staff at the firm understand that none of them may accept a prompt to update any AI software without this person’s prior assent.

 

IV.  CONCLUSION

 GenAI offers an unprecedented opportunity for solo and small-firm practitioners to streamline administrative burdens and deliver cost-effective representation. However, New York regulatory authorities, federal evidentiary rulings, and appellate courts have made it clear that technology represents an efficiency multiplier, not an ethical exemption. Under NYRPC 3.1, 22 NYCRR Part 161, Part 130, and the developing common law of privilege waiver, an attorney’s signature remains an absolute, non-delegable personal guarantee of accuracy, confidentiality, and systemic integrity. By implementing the protocol outlined in this opinion, the small-firm bar can safely navigate the personal sanctions risks and effectively insulate their practices and their clients’ privileges from catastrophic exposure. Attorneys should continue to monitor changes in the AI platforms which they use in order to stay abreast of this rapidly changing technology and should routinely ensure that their AI-specific protocols are current, effective, and comply with ethical rules.

Issued by:

New York County Lawyers Association – Committee on Professional Ethics