State Bar of New Mexico Formal Ethics Advisory Opinion 2024-004 — Using Generative Artificial Intelligence in the Practice of Law

Formal Opinion 2024-004, "Using Generative Artificial Intelligence in the Practice of Law," is a formal ethics advisory opinion of the Ethics Advisory Committee of the State Bar of New Mexico. It was issued on September 24, 2024 and republished in the State Bar's Bar Bulletin, Volume 63, No. 11-D, on November 27, 2024. It answers a single question, whether a lawyer may use generative artificial intelligence in the practice of law, in the affirmative, and sets out the duties under the New Mexico Rules of Professional Conduct that such use engages: competence, diligence, client communication, reasonable fees, confidentiality, duties to former clients, candor toward the tribunal, truthfulness, and supervision.

Its central requirement for court filings is that a lawyer using a generative AI tool to draft a document submitted to a court "must independently verify the accuracy and sufficiency of all research, citations, and analysis conducted by the AI tool," and may not rely on the same tool to check its own output. It closes by stating that a New Mexico lawyer who chooses to use generative AI "must do so responsibly" and must adhere to both the Rules of Professional Conduct and "any tribunal or agency rules, policies, procedures or orders."

Status

The opinion states its own force in a disclaimer printed at its head. The Committee "is constituted for the purpose of advising lawyers on the application of the New Mexico Rules of Professional Conduct," and its opinions "are advisory only, and are not binding on lawyers, the disciplinary board, or any tribunal." Conclusions rest on the facts referenced in the opinion and on the Rules in effect on the date of issue, and lawyers are cautioned that "should the Rules subsequently be revised, or different facts be presented, a different conclusion may be appropriate." The Committee states that it does not opine on matters of substantive law, and that the statements in an opinion are "the consensus of the Committee members who considered the question(s) presented." No member or chair is named.

The opinion cites the Rules as NMRA (2024). It is guidance on how those Rules apply, and by its own terms is neither a rule nor a determination binding on the disciplinary board.

Question presented and summary answer

The question presented is "May a lawyer use Generative Artificial Intelligence in the practice of law?" The summary answer reads: "Under the New Mexico Rules of Professional Conduct, yes, they may; however, use of such technology raises a number of considerations which are highlighted here."

The analysis defines artificial intelligence by quotation from Merriam-Webster and distinguishes generative AI, which "is capable of generating content" such as correspondence, retention agreements, briefs, contracts, case summaries and estate planning documents, from the non-generative AI that lawyers "have historically used," primarily in legal research and document assembly. It adds that generative AI can analyze data, records and legal documents, search for relevant precedent, and suggest edits to existing materials.

Competence, diligence and the terms of permitted use

The Committee states that "the responsible use of Generative AI is consistent with lawyers' duties of competence (Rule 16-101, NMRA), diligence (Rule 16-103, NMRA) and charging reasonable fees and costs to clients (Rule 16-105, NMRA)." It quotes Comment 9 to Rule 16-101, under which a lawyer should keep abreast of changes in the law and its practice "including the benefits and risks associated with relevant technology," adding emphasis to that phrase. It lists potential benefits: more efficient initial drafting and routine correspondence, and so lower cost to clients; help in considering alternative approaches to drafting complex contracts or cross-examining witnesses; and time and cost savings in discovery.

The opinion does not require use of the technology: "The Committee is not suggesting that, currently, lawyers must use Generative AI in the practice of law." Lawyers who do use it "must do so responsibly, recognizing that the use of Generative AI does not change their fundamental duties under the Rules of Professional Conduct." Three conditions follow. Use must be tailored to the lawyer's own practice, to each matter, and to each client. The lawyer should determine whether a matter requires discussing AI use with the client under Rule 16-104(A)(2). And the lawyer "must be familiar with" the rules, policies or procedures of any court, agency or administrative body on generative AI, "including whether it is allowed and what level of disclosure is required if it is used."

Rules construed

The opinion lists ten rules as implicated. It names five areas as "most commonly implicated" and treats them in numbered sections "in numerical order for ease of reference," followed by a separate consideration on client communication.

Confidentiality (Rule 16-106)

The opinion identifies three features of generative AI tools that put client confidentiality at risk: many are owned by third-party providers "with varying degrees of security and policies regarding sharing data"; they may require prompts containing specific information about a case; and many "are self-learning and therefore keep and use the information with which they are prompted to train the AI for future users." It relies on Comment 4 to Rule 16-106 for the principle that a lawyer must not reveal information relating to the representation absent informed consent, on Comment 6 for its extension to disclosures that "could reasonably lead to the discovery" of protected information, and on Rule 16-106(C)'s requirement to "make reasonable efforts to prevent the inadvertent... disclosure of, or the unauthorized access to, information relating to the representation of a client."

From these it derives four steps. First, lawyers must understand a tool's parameters and policies, including whether it shares data with third parties, stores and uses data to train its own AI, and has adequate security against unintended access. Second, they must not enter confidential information, or information that could lead to its discovery, into tools lacking adequate confidentiality and security protections, a category that "includes Generative AI tools that share inputted information with third parties or keep inputted information to train their AIs." Third, they "should always anonymize client information" and avoid details that could identify the client. Fourth, "in appropriate cases" they should obtain the client's informed consent before entering confidential information.

Candor toward the tribunal and truthfulness (Rules 16-303 and 16-804(C))

The opinion states that using generative AI to draft court filings "directly implicates" the duties of candor and truthfulness, because the tools "can create inaccurate or false outputs either caused by AI hallucinations or by incomplete and careless prompts by the attorney using the tool." It quotes Rule 16-303(A)(1), under which a lawyer shall not knowingly "make a false statement of fact or law to a tribunal or fail to correct a false statement of material fact or law previously made to the tribunal by the lawyer"; Comment 2's statement that a "lawyer must not allow the tribunal to be misled by false statements of law or fact or evidence that the lawyer knows to be false"; Comment 4 on legal argument "based on a knowingly false representation of law"; and Rule 16-804(C) on conduct involving "dishonesty, fraud, deceit or misrepresentation."

On verification it sets four requirements:

1. A lawyer "must independently verify the accuracy and sufficiency of all research, citations, and analysis conducted by the AI tool."
2. The lawyer "must not entirely rely on the Generative AI tool used to draft the document in the first place to verify the accuracy of its research or analysis," because such methods "have been shown to be ineffective at catching mistakes made by the Generative AI tool as such tools can be misled by their own prior hallucination." The authority given is Mata v. Avianca, Inc., 678 F. Supp. 3d 443, 458 (S.D.N.Y. 2023), described as discussing a lawyer's failure "to independently verify the accuracy and existence of the sources cited by the tool used."
3. The lawyer "must correct any errors or mistakes made by the tool." The opinion states that "[f]ailure to do so is a knowing misrepresentation of the law and a violation of, among other rules, the lawyer's duty of candor to the court and the lawyer's duty to avoid engaging in dishonesty or misrepresentation."
4. The lawyer "should research and abide by" the rules, orders or other requirements of the relevant jurisdiction on the use and disclosure of generative AI. The opinion cites the April 6, 2024 report of the New York State Bar Association Task Force on A.I. for the statement that many courts are adopting such requirements, and states that abiding by them "also falls under a lawyer's duty of candor to the court."

Fees and costs (Rule 16-105)

Because generative AI can make research, drafting and preparation for proceedings more efficient, the opinion treats it as affecting "the time and labor involved in the representation of clients" and so the reasonableness of fees under Rule 16-105(A)(1). A lawyer "must be conscientious of the actual time spent on a given task and only charge the client for the time and labor actually incurred." Where the tool makes a task more efficient, "the lawyer must only use the time the lawyer actually spent on the given task to calculate their fee and not include the time saved by the use of the tool." Costs of using a tool may be passed on, but the lawyer "must inform the client of the actual cost incurred from using the tool in writing and the lawyer's intent to charge the client that cost prior to the representation." Drawing on Comments 1 and 2 to Rule 16-105, the opinion states that costs must be agreed in advance or reasonably reflect costs actually incurred, and describes a detailed engagement letter covering scope, fee basis and costs as "best practices."

Duties to former clients (Rule 16-109)

The Committee states that, while a lawyer should consider all potential conflicts, it regards the duty to former clients as the most prevalent conflict-of-interest risk in using generative AI. Because tools "often store and use prior prompts and outputs to train and improve" themselves, future outputs may use information from a prior representation, or from another lawyer's concurrent representation in the same firm, to that client's disadvantage, which engages Rule 16-109(C)(1)–(2). A lawyer must know whether the tool trains on prior inputs and what prior client information has been entered into it. If the tool lacks safeguards to protect prior client information and screen potential conflicts, "the lawyer must verify independently that no conflict exists with prior client information inputted into the AI," and "[i]f the lawyer cannot independently verify a lack of conflict, then the lawyer should not use the tool for their current representation." Lawyers should also avoid entering confidential information that could create a conflict with their own future clients or their firm's other clients.

Supervision (Rules 16-501 and 16-503)

Partners and lawyers with managerial authority must make reasonable efforts to ensure that the firm's lawyers and nonlawyer assistants conform to the Rules, including through internal policies, instruction and supervision. Applied to generative AI, the opinion states that "prior to using Generative AI, lawyers must develop appropriate policies and procedures on the ethical use of Generative AI and train all other lawyers and staff in the law firm on such policies and procedures," and should periodically review compliance. Failure to do so "could expose a lawyer to liability for the misuse of Generative AI in violation of the Rules of Professional Conduct by other lawyers or legal staff in the lawyer's law firm."

Client communication (Rule 16-104)

In a closing section headed "An Additional Consideration," the opinion states that whether a lawyer must discuss the risks and benefits of generative AI with a client "depends upon a number of factors," and that it is "not possible to identify or illustrate every situation." It names three circumstances in which lawyers "may need to discuss their use of Generative AI with a client": where its use "plays a significant role in producing final products for clients"; where "a client inquires about a lawyer's use of Generative AI"; and where its use "requires the lawyer to disclose confidential client data to obtain any resulting product." Other circumstances may also require informing the client, and each matter is to be evaluated "on a case-by-case basis."

The opinion states no general duty to disclose AI use to clients. On disclosure to courts, its only statement is the requirement to learn and follow each tribunal's own rules.

Matters the opinion does not address

• Factual content in filings. The verification requirement is stated for "research, citations, and analysis conducted by the AI tool," and the one case cited concerns sources that did not exist. The opinion does not separately discuss AI-generated statements of the factual record, such as summaries of trial testimony, although the text of Rule 16-303(A)(1) it quotes covers false statements "of fact or law."
• Witness testimony. The only reference to witnesses is the listed benefit that generative AI "can assist a lawyer in considering alternative approaches to drafting complex contracts or cross-examining witnesses." The opinion does not address use of generative AI in preparing, summarizing or characterizing witness testimony.
• Named products. No tool, vendor or model is named.
• Consequences. The opinion characterizes a failure to correct AI errors as "a knowing misrepresentation of the law" and a rule violation, but does not discuss sanctions or discipline.
• Other guidance. No American Bar Association opinion and no other state bar's opinion is cited.

Relation to the Aarons contempt order

The New Mexico Supreme Court's 2026 order holding attorney Stephen D. Aarons in direct contempt (In re Stephen D. Aarons — Dispositional Order of Direct Contempt (N.M. Sup. Ct., 2026)) concerns the conduct the opinion's candor section addresses: Aarons admitted using ChatGPT to prepare a brief in chief and not verifying its factual claims or legal authority before signing and filing it. The order does not cite Formal Opinion 2024-004 or any Rule of Professional Conduct. It grounds the contempt finding in Rule 12-312(D) NMRA, under which failure to comply with the briefing requirements of Rule 12-318 NMRA is grounds for sanctions, and in the Court's superintending authority under Article VI, Section 3 of the New Mexico Constitution, and it refers the matter to the Disciplinary Board, a body the opinion names as not bound by the Committee's opinions.

The defects the order records differ in kind from the example the opinion gives. Four are wholly fabricated witnesses and three are false testimony attributed to real witnesses, alongside two misrepresented New Mexico authorities; the opinion's verification language addresses research, citations and analysis, and its cited example concerns non-existent sources. The order also records that Aarons did not tell his client the brief was unverified and contained misrepresentations; the opinion's client-communication section addresses when the use of AI should be discussed with a client, and does not address disclosure of errors.

Reception

In Dehghani v. Castro, No. 2:25-cv-00052-MIS-DLM (D.N.M. May 9, 2025), United States District Judge Margaret Strickland quoted the opinion in affirming a sanctions order. Magistrate Judge Damian L. Martinez had found on April 2, 2025 that attorney Felipe D.J. Millan violated Federal Rule of Civil Procedure 11(b) by filing a brief, written by a freelance attorney engaged through the company LAWCLERK, in which Millan himself identified six cited cases that did not exist and thirteen citations to cases that did not support the propositions stated. The sanctions were a $1,500 fine payable to the court; forwarding the order to LAWCLERK; completion of a one-hour CLE-credited program "related to legal ethics in writing or the use of AI in writing"; self-reports to the New Mexico and Texas state bar disciplinary boards; and a report of the contract attorney to the New York state bar disciplinary board. Millan's objection was directed at the self-reporting requirement; he also asked the court to reconsider the fine but made no argument on it and had already paid it.

In overruling the objection, the district court quoted the opinion's statement that a lawyer "must pay particular care" to Rules 16-303 and 16-804(C) when using generative AI to draft court filings and its requirement to "independently verify the accuracy and sufficiency of all research, citations, and analysis conducted by the AI tool," citing it as "Formal Opinion 2024-004, at 5." It also quoted Opinion No. 705 (February 2025) of the Professional Ethics Committee of the State Bar of Texas. The court described both as issued by "[t]he ethics disciplinary bodies in both New Mexico and Texas" and treated their reasoning as "instructive," while noting that "it is not certain that AI was in fact used here" and that Millan "did not himself utilize AI" (Source: caselaw.findlaw.com). The court's description of the issuing body as disciplinary differs from the opinion's own disclaimer, which describes the Committee as advisory and its opinions as not binding on the disciplinary board.

Provenance

The text was captured on 2026-09-16 from the Bar Bulletin PDF on the State Bar of New Mexico's own domain, where the opinion occupies pages 13–15 of a 54-page issue. The ingest task and its verification record had estimated pages 13–16 from the table of contents; page 16 is a full-page call for Bar Bulletin article submissions. The Committee also publishes the opinion as a standalone nine-page PDF on the same host, generated September 30, 2024, which is the version the District of New Mexico cited. The two texts are word-for-word identical apart from the standalone's header block, colons after two headings, and a misspelling in one heading ("CONDCUT") that the Bulletin corrects. Pin cites of the form "at 5" refer to the standalone pagination.

Confidence is medium. The full text was read and is corroborated by two copies from the issuing body and by a federal court's quotation, but whether the Committee has revised or withdrawn the opinion since 2024 was not checked.

Relationships

• supports: AI-Fabricated Citations — a professional-responsibility standard requiring independent verification of AI-generated research, citations and analysis before filing
• supports: Legal Services — AI Deployment — bar guidance on generative AI covering confidentiality, candor, fees, conflicts, supervision and client communication
• related: In re Stephen D. Aarons — Dispositional Order of Direct Contempt (N.M. Sup. Ct., 2026) — a 2026 New Mexico Supreme Court contempt order on an unverified ChatGPT-assisted brief; the order does not cite the opinion
• related: State Bar of New Mexico — the issuing body, through its Ethics Advisory Committee
• related: Sycophancy and Hallucination — the opinion grounds verification in hallucination and in tools being "misled by their own prior hallucination"
• related: AI Liability — responsibility for unverified AI output placed on the filing lawyer, and supervisory exposure on firm managers
• related: AI and Privacy — confidentiality risk from tools that retain prompts and train on them
• related: China — SPC Opinions on Lawfully Adjudicating AI-Related Dispute Cases (法发〔2026〕10号) — China's 2026 judicial rule imposing verification and disclosure duties on AI-assisted filings; the New Mexico opinion states a verification duty and leaves court disclosure to each tribunal's rules

Sources

• Raw source: Raw Sources/State Bar of New Mexico - Formal Ethics Advisory Opinion 2024-004 on Using Generative AI in the Practice of Law (Bar Bulletin Vol. 63 No. 11-D, 2024-11-27).md
• Bar Bulletin, Vol. 63, No. 11-D, November 27, 2024, pp. 13–15: sbnm.org
• Standalone opinion PDF (State Bar of New Mexico): sbnm.org
• Dehghani v. Castro, No. 2:25-cv-00052-MIS-DLM (D.N.M. May 9, 2025) (Source: caselaw.findlaw.com)