ABA Formal Opinion 512: Lawyers' Use of Generative AI and Confidentiality, Supervision and Fees

On July 29, 2024, the American Bar Association (ABA) Standing Committee on Ethics and Professional Responsibility published Formal Opinion 512, "Generative Artificial Intelligence Tools", a comprehensive analysis of the professional duties of lawyers who use generative AI tools. The opinion sets out interpretations tied to specific provisions of the ABA Model Rules of Professional Conduct on six issues: competence, confidentiality, communication with clients, candor toward the tribunal, supervisory responsibilities, and legal fees. As of the time of writing (July 12, 2026), about two years have passed since its publication, and during that period several U.S. state bars have also moved to publish their own opinions on the same subject. This article explains the content of the opinion published in July 2024 together with its legal nature, and then converts it into items that Japanese attorneys and in-house legal staff can compare against their own practice.

Legal Nature and Scope of Formal Opinion 512

Formal Opinion 512 is an opinion issued by the ABA Standing Committee on Ethics and Professional Responsibility interpreting the Model Rules of Professional Conduct as amended through August 2023. The Model Rules are model rules drafted by the ABA and are not, in themselves, legal norms that directly bind lawyers. In the United States, each state's supreme court or state bar adopts its own rules of professional conduct based on the Model Rules, and it is the rules adopted by that state that actually bind lawyers. Formal Opinion 512 is therefore a committee view that may be referred to when interpreting each state's rules; it does not take precedence over, or replace, the state rules themselves. The opinion itself also states the expectation that the Committee and state and local bar ethics committees will continue to update their guidance as GAI tools develop, indicating that Formal Opinion 512 is not fixed as the only interpretive framework.

Its scope also requires attention. The opinion does not address AI technology in general, such as the AI already used in electronic discovery or contract analysis in M&A, but focuses on generative AI (GAI), which creates new content such as text and images. The Model Rules it addresses are Rule 1.1 on competence, Rule 1.6 on confidentiality, Rule 1.4 on communication with clients, Rules 3.1, 3.3 and 8.4(c) on candor toward the tribunal, Rules 5.1 and 5.3 on supervisory responsibilities, and Rule 1.5 on legal fees.

Competence, Confidentiality and Communication with Clients

On Rule 1.1 (competence), the opinion states that lawyers need not become GAI experts; it is sufficient to have a reasonable understanding of the benefits and risks of the GAI tools they use. How far the output of GAI must be independently verified, however, depends on the tool and the task. In an example given by the opinion, where a lawyer uses GAI to summarize a large number of contracts, if the lawyer has first reviewed some of the documents manually and checked the accuracy by comparing them with the summaries, the lawyer need not thereafter manually recheck every document.

On Rule 1.6 (confidentiality), the opinion points out that if information relating to the representation of a client is entered into a self-learning GAI tool, that information may be reflected in output given to a lawyer handling another client's matter, and the client's confidential information may leak inside or outside the firm. For this reason, before such information is entered into a self-learning GAI tool, the client's informed consent must be obtained, and simply placing boilerplate language permitting GAI use in an engagement letter is not sufficient. This conclusion is premised on the risks and capabilities of GAI tools at the time of publication, and the opinion expressly states that it may change as technology develops; this should be kept in mind when reading it now, about two years after publication. Because vendors' data retention policies and whether data is used for training differ according to the terms of each GAI service, I think it is necessary to recheck the terms of use before applying the opinion's conclusion directly to current tools.

On Rule 1.4 (communication with clients), disclosure or consultation is required where the client asks whether GAI is used, where GAI use bears on the reasonableness of fees, or where GAI output influences a significant decision in the representation.

Candor Toward the Tribunal and Supervisory Responsibilities

On Rule 3.1 (meritorious claims), Rule 3.3 (candor toward the tribunal) and Rule 8.4(c) (prohibition of misconduct), the opinion explains, in light of the risk of "hallucination" in which GAI presents case law that does not exist, the duty to check output and citations before filing with a court and to correct errors. The opinion also notes that there have already been cases in which lawyers were sanctioned after filing erroneous GAI output without verification, but the course of individual cases is outside the scope of this article. The importance of not taking output at face value and always having a human perform the final check is also covered in Do Not Use Generative AI Answers As-Is: Why a Human Must Always Do the Final Check.

On Rules 5.1 and 5.3 (supervisory responsibilities), the opinion states that managerial lawyers at a firm must establish policies on GAI use, and supervisory lawyers must train subordinate lawyers and nonlawyer staff on the capabilities and limitations of GAI, including data handling. Where an external GAI vendor is used, the opinion refers to existing opinions on cloud services and outsourcing and states that lawyers should check whether confidentiality settings are effective, whether there is a notification system in case of a data breach, whether data is retained or rights are asserted over it after the contract ends, and the risk of cyberattacks.

Legal Fees

On Rule 1.5 (legal fees), the opinion states that where a lawyer is engaged on an hourly basis, even if GAI use makes the work more efficient, the lawyer may not bill for more than the time actually spent. For flat fees or contingency fees as well, the opinion states that maintaining the same fee as without GAI for work that GAI allowed to be completed in significantly less time than before may violate the reasonableness requirement of Rule 1.5. As for the treatment of the cost of GAI tools, the opinion states that the cost of general-purpose tools, such as a grammar-checking function built into word-processing software, should be treated as the firm's overhead, while the actual cost of using a pay-per-use third-party GAI service to review a large volume of contracts may be charged as an expense with the client's consent. In addition, the time a lawyer spends learning GAI tools used in ordinary practice cannot be billed to the client, but where the client specifies the use of a particular unfamiliar tool, the learning time may be billed, subject to prior agreement.

Items for Japanese Attorneys and Corporate Legal Teams to Check

The Model Rules of Professional Conduct are model rules adopted by individual U.S. states, and Formal Opinion 512 is an opinion interpreting them. Japanese attorneys are governed by the Attorney Act and the Basic Rules on the Duties of Practicing Attorneys, and the statements in the opinion cannot be read directly as obligations under Japanese law. There are, however, comparable items. Article 7 (self-improvement) of the Basic Rules overlaps with the approach of Rule 1.1 in that it imposes on attorneys a duty to endeavor to be well versed in laws and legal affairs, but there is no express provision requiring technical competence specific to GAI tools. Article 23 (confidentiality) of the Basic Rules provides that attorneys must not disclose or use a client's confidential information, but there is no express provision requiring the client's consent when entering information into GAI. Article 19 of the Basic Rules provides for guidance and supervision of office staff and others, but unlike Model Rule 5.1, it does not set out in a separate provision a system for supervising GAI use by the firm's own attorneys. Article 24 (attorney fees) and Article 29 (explanations upon accepting a matter) of the Basic Rules provide for the appropriateness of fees and the duty to explain when accepting a matter, and there is room to consider them alongside the Rule 1.5 issue of how efficiencies from GAI use should be reflected in fees.

When engaging an outside law firm, in-house legal staff may add the following to the items to check: what consent procedures and usage policies the firm has in place for GAI tools into which matter information is entered, whether billing under hourly arrangements reflects actual time worked, and whether flat-fee quotes have been revised on the premise of GAI use. The design of an in-house generative AI use policy is explained in How to Create a Generative AI Use Policy, and the division of roles between AI and attorneys in What Is an AI Lawyer? How to Divide Work Between Attorneys and AI in Corporate Legal Practice in the Generative AI Era. The point of checking article numbers and sources yourself rather than using them as output is also shared with Always Check Article Numbers and Sources Yourself: A Pitfall of Generative AI.

Developments to Watch

Since Formal Opinion 512 was published, similar opinions have followed from U.S. state bars. As one example that could be confirmed, Opinion 705, published in February 2025 by the Professional Ethics Committee for the State Bar of Texas, sets out an analysis under the Texas rules of professional conduct within a similar framework of competence, confidentiality, verification of output, and reasonableness of fees. Other states have also published their own opinions and guidance since 2024, but their overall picture and the details of each opinion could not be confirmed as of the time of writing.

Whether the ABA itself has published an opinion revising or supplementing Formal Opinion 512 also could not be confirmed as of the time of writing. The terms on which GAI tools are offered have continued to change since the opinion was published, and the opinion itself positions its conclusions, particularly on confidentiality, as provisional and premised on the tools available at the time of publication. When Japanese in-house legal staff engage U.S. lawyers, I think the practical starting point is not to rely on the statements in Formal Opinion 512 as they are, but to check the contractual terms of the GAI tools the firm currently uses and the latest treatment under the rules of professional conduct of the state to which the firm belongs.