ABA Formal Opinion 512 (2024): Generative AI Duties for Lawyers

Everyone remembers Mata v. Avianca for the wrong reason.

In June 2023, Judge P. Kevin Castel of the Southern District of New York imposed a $5,000 penalty, jointly and severally, on two lawyers and their firm after they filed a brief citing court opinions that ChatGPT had invented (Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), Opinion and Order on Sanctions, filed June 22, 2023). The sanctions were imposed under Rule 11 or, alternatively, the court's inherent authority. One of the lawyers, Steven Schwartz, later filed screenshots of his questions to the same chatbot, including whether Varghese was a real case, and it answered that the authorities were real. The court did not accept that as an excuse. It noted that the screenshots were undated and that his affidavit "strongly suggested" he had doubted Varghese even before the filings. It found bad faith on the part of the two individual lawyers, "based upon acts of conscious avoidance and false and misleading statements to the Court," and noted that they "doubled down" after the court's orders called the cases into question.

Reviewed and corrected on 12 October 2026. This is general information, not legal advice; check the cited opinion or rule before relying on it.

The story went viral because it was embarrassing, and the lesson everyone took home was simple: check your citations.

That lesson is correct, and it is also the smallest part of what the ABA later wrote. When the ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512 on generative artificial intelligence tools on July 29, 2024, it did not write a hallucination rule. It walked through existing duties and explained how each one applies when a lawyer uses a generative AI tool.

The hallucination problem is covered. So are other obligations that apply even when your AI never invents a single case and nothing gets filed: confidentiality, communication, supervision and fees.

Short answer: ABA Formal Opinion 512, issued July 29, 2024 by the Standing Committee on Ethics and Professional Responsibility, applies existing Model Rules to generative AI. It covers competence (1.1), confidentiality (1.6), communication (1.4), meritorious claims and candor (3.1, 3.3, 8.4(c)), supervision (5.1, 5.3), and fees (1.5). In practice that means understanding the tool, independently verifying its output to an appropriate degree, getting informed consent before client information goes into a self-learning tool, and billing hourly clients only for the time you actually spend.

TL;DR

• ABA Formal Opinion 512 (July 29, 2024) applies existing Model Rules to generative AI: competence (1.1), confidentiality (1.6), communication (1.4), meritorious claims and candor (3.1, 3.3, 8.4(c)), supervision (5.1/5.3), and fees (1.5).
• The opinion declines to set a fixed verification standard. The appropriate amount of review "will necessarily depend on the GAI tool and the specific task that it performs." There is no ABA checklist that makes you compliant.
• Two duties get less attention than citation checking: fees and confidentiality. You may not charge a client to learn a tool you will regularly use, and you need informed consent before putting client information into a self-learning tool.
• Mata v. Avianca is the best-known sanctions case, and 512 does not cite it. The opinion does note that problems so far include "citations to nonexistent opinions."
• Adoption is still early: in the ABA's 2024 Legal Technology Survey Report, 30% of respondents reported using AI, and accuracy was the top concern at 75% (figures from a March 2025 LawSites summary of the report).

Official text, as of 11 October 2026

• Source: ABA Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512, "Generative Artificial Intelligence Tools," July 29, 2024. Read the opinion (PDF).
• What it says on verification: relying on or submitting a tool's output without an appropriate degree of independent verification or review "could violate the duty to provide competent representation as required by Model Rule 1.1."
• What it is: guidance. The opinion says it "offers general guidance," and it is based on the Model Rules as amended through August 2023.
• What changed in 2026: the text of 512 above is the July 2024 opinion. The movement since then is at the state level. On May 14, 2026, California's Board of Trustees approved updated revisions to the State Bar's Practical Guidance on generative AI, which replaces the version first published on November 16, 2023 and addresses agentic AI. Details are in the state table below.

Part of our legal AI verification and hallucination guide series.

For related verification / hallucination / vendor-trust coverage, see Are ABA Formal Opinions Binding? What Opinion 512 Means for Your AI Workflow and How to Verify AI Legal Citations Before You File (ABA 512 Checklist).

Why 512 Had to Exist

The ABA does not issue a formal opinion because a technology is new. It issues one when lawyers have questions the existing rules do not answer on their face.

By mid-2024, Mata was no longer a one-off. Lawyers were filing AI-supplied citations that did not exist. 512 does not cite Mata or any sanctions case. In section D it says only that "issues that have arisen to date with lawyers' use of GAI outputs include citations to nonexistent opinions, inaccurate analysis of authority, and use of misleading arguments," and it points to D.C. Bar Opinion 388 (2024) in a footnote. Texas Opinion 705, issued in February 2025, does cite Mata and calls it "the most famous example."

The court in Mata was careful to say there is "nothing inherently improper about using a reliable artificial intelligence tool for assistance." What it objected to was filing the output without checking it, and then standing by it. We cover that case and the verification protocol in our piece on AI hallucinations and sanctions, so we will not repeat it here.

The full opinion is on americanbar.org (link above). A readable walkthrough is Wendy J. Muchman's article in the National Conference of Bar Examiners' Bar Examiner, article on Opinion 512 (Fall 2024). Muchman is listed on the Standing Committee roster in the opinion.

512 treats generative AI as something a lawyer remains fully responsible for. In its words, "regardless of the level of review the lawyer selects, the lawyer is fully responsible for the work on behalf of the client." The opinion also says lawyers "may not abdicate their responsibilities by relying solely on a GAI tool to perform tasks that call for the exercise of professional judgment."

A useful way to think about this (the author's framing, not the opinion's) is to treat the tool like a fast helper that can be confidently wrong. You would not file that helper's work without reading it, and you would not let it decide what to tell the client.

The opinion's discussion has six lettered sections (competence, confidentiality, communication, meritorious claims and candor, supervision, fees). Its conclusion lists the duties slightly differently, splitting out items such as candor and meritorious claims.

Rule 1.1 (Competence): You Have to Understand the Tool, Not Just Use It

Competence under 512 has two layers, and most coverage only mentions the first.

The first is technological competence. Comment [8] to Rule 1.1 refers to "the benefits and risks associated with relevant technology," and 512 applies it to generative AI. Lawyers "need not become GAI experts," but they need "a reasonable understanding of the capabilities and limitations of the specific GAI technology" they use, or they should draw on the expertise of others who do.

The second layer turns competence into a verification duty. Because these tools can produce inaccurate output, 512 says relying on or submitting that output without an appropriate degree of independent verification or review "could violate" Rule 1.1. 512 does not say how much verification is enough. The required level of review "will necessarily depend on the GAI tool and the specific task that it performs."

The opinion gives two examples. A lawyer who has tested a contract-summarizing tool on a smaller set of documents, compared its summaries to the originals and found them accurate would not necessarily have to review every document by hand. And a tool "designed specifically for the practice of law or to perform a discrete legal task" may require less independent verification, "particularly where a lawyer's prior experience with the GAI tool provides a reasonable basis for relying on its results."

That makes this a tool-selection and workflow problem. The ABA will not give you a number, so the burden lands on the process you build around the tool.

One practical point (our view, not 512's): a tool that shows you the source it relied on lets you open the citation, read the passage and confirm it says what the tool claims. A tool that produces a citation with no source forces you to redo the research. Grounding lowers the cost of checking. It does not remove the need to check. 512 itself cites, in footnote 14, a Stanford study finding that leading legal research tools "hallucinate between 17% and 33% of the time." (How retrieval changes the failure modes is its own topic, covered in how AI legal research works.)

Rule 1.6 (Confidentiality): The Duty Most Lawyers Underestimate

Here is the first rule that applies even when nothing hallucinates and nothing gets filed.

512 addresses what happens when you type client information into a generative AI tool. It describes "self-learning" tools as ones that "will learn from themselves as they cull more data." The risk it identifies is that information about one client's matter goes into the tool and is later revealed in response to prompts from other users, or from people inside the same firm who are walled off from the matter. Its holding: because many self-learning tools are designed so that their output could lead directly or indirectly to disclosure, "a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool."

Two words do a lot of work.

"Self-learning" matters because the consent requirement is tied to that risk. The opinion also says that the analysis "will be fact-driven and depend on the client, the matter, the task, and the GAI tool." A tool that does not learn from your inputs presents a different risk than one that does, but you still have to assess it. The opinion says that if you are using a tool for something like idea generation without inputting information relating to the representation, informed consent is not necessary.

"Informed consent" matters because 512 says what it takes. The client must have the lawyer's best judgment on why the tool is being used, the extent of and specific information about the risk (including the kinds of client information that will be disclosed and the ways others might use it against the client's interests), and a clear explanation of the tool's benefits. And "merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient."

The practical defense is to know where your inputs go before you type them. 512 says lawyers should "read and understand the Terms of Use, privacy policy, and related contractual terms and policies of any GAI tool they use," or consult a colleague or outside expert who has. The longer treatment of vendor posture lives in where your legal AI data actually goes.

Rule 1.4 (Communication): When You Have to Tell the Client

512 starts from Rule 1.6. Where 1.6 does not require disclosure and informed consent, the lawyer "must separately consider" whether Rule 1.4 requires disclosing the use of a tool.

The opinion does not impose a blanket duty to disclose every use of AI. It says "depending on the circumstances, client disclosure may be unnecessary." But it lists situations where disclosure is required:

• The client asks how the work was done, or whether AI was used.
• The client requires disclosure in the engagement agreement or in outside counsel guidelines.
• The lawyer needs the client's informed consent to input client information (see Rule 1.6 above).
• The use of the tool is relevant to the basis or reasonableness of the lawyer's fee.
• The tool's output will influence a significant decision in the representation, such as when a lawyer relies on AI to evaluate potential litigation outcomes or jury selection.

512 also says it is not possible to catalogue every situation, and that even when disclosure is not required, lawyers may tell clients how they use AI. The engagement agreement is "a logical place" for that.

When you do need Rule 1.6 consent, that conversation also tells the client about the AI use, so there is no reason to have it twice. Where 1.6 does not require consent, ask separately whether 1.4 requires disclosure.

Rules 3.1, 3.3 and 8.4(c) (Candor) and Rules 5.1/5.3 (Supervision)

These are the rules everyone already associates with AI, so this section is brief.

Meritorious claims and candor. 512 says Rules 3.1, 3.3 and 8.4(c) "may be implicated by certain uses" of generative AI. Rule 3.3 covers a lawyer who "knowingly" makes a false statement of fact or law to a tribunal. Rule 8.4(c) bars conduct involving dishonesty, fraud, deceit or misrepresentation, and 512 adds that "even an unintentional misstatement to a court can involve a misrepresentation under Rule 8.4(c)." So AI output "must be carefully reviewed to ensure that the assertions made to the court are not false." Before submitting materials, lawyers must review AI output, "including analysis and citations to authority," and correct errors.

Mata itself was a Rule 11 sanction case, and Rule 3.3 covers similar conduct. Schwartz and LoDuca were sanctioned under Rule 11(b)(2) for filing the fake cases and then standing by them. The court's finding was that of subjective bad faith, which is a higher bar than carelessness.

512 adds two practical notes for litigators. Some courts "have responded by requiring lawyers to disclose their use of GAI," so check the local rules of the court you are in. And, depending on the facts, Rule 3.4(c) may also be in play.

The verification protocol that satisfies these rules (what to check, in what order, with what evidence) lives in our sanctions deep-dive.

Supervision (5.1 and 5.3). 512 says "managerial lawyers must establish clear policies regarding the law firm's permissible use of GAI, and supervisory lawyers must make reasonable efforts to ensure that the firm's lawyers and nonlawyers comply with their professional obligations when using GAI tools." Supervisors must also make sure subordinate lawyers and nonlawyers are trained on the ethical and practical use of the tools and the risks of using them. A written policy is the practical way to do that, though 512 itself says "clear policies" and does not use the word "written."

512 also applies Rule 5.3(b) to vendors: lawyers have "additional supervisory obligations insofar as they rely on others outside the law firm to employ GAI tools," and "these concepts also apply to GAI providers and tools."

512 does not describe the AI tool as a nonlawyer assistant. That comparison comes from state opinions. Florida Opinion 24-1 says "many of the standards applicable to nonlawyer assistants provide useful guidance for a lawyer's use of generative AI," and Texas Opinion 705 says lawyers who rely on generative AI risk "many of the same perils as those who rely on inexperienced or overconfident nonlawyer assistants."

In short, as our summary rather than a quotation from any opinion: the tool is not a lawyer, you are the supervising lawyer, and the signature on the filing is yours.

Rule 1.5 (Fees): What 512 Says About Billing for AI

Fees get less attention than hallucinations, and in daily practice they may be where lawyers are most exposed.

512 makes several points about fees.

You may not charge a client for your learning curve. 512 says "a lawyer may not charge a client to learn about how to use a GAI tool or service that the lawyer will regularly use for clients because lawyers must maintain competence in the tools they use." There is an exception: "if a client explicitly requests that a specific GAI tool be used in furtherance of the matter and the lawyer is not knowledgeable in using that tool, it may be appropriate for the lawyer to bill the client to gain the knowledge to use the tool effectively." The lawyer and client should agree on any new billing terms first, preferably in writing.

On hourly matters, bill actual time. Lawyers who bill by the hour "must bill for their actual time." 512 relies on ABA Formal Opinion 93-379: if a lawyer is particularly efficient, "it nonetheless will not be permissible to charge the client for more hours than were actually expended." So if a task that took three hours now takes twenty minutes, you cannot bill three hours. You can bill the time you spend entering information and the time you spend reviewing the output for accuracy and completeness.

On flat and contingent fees, the Rule 1.5(a) factors still apply. 512 says that if a tool lets a lawyer finish tasks much more quickly, "it may be unreasonable under Rule 1.5 for the lawyer to charge the same flat fee when using the GAI tool as when not using it."

Charging for the tool itself. Before charging a client for AI tools or services, 512 says, "the lawyer must explain the basis for the charge, preferably in writing" (Rule 1.5(b)). It then distinguishes by type of tool. A tool that functions like equipping and maintaining a practice, such as a grammar checker embedded in word processing software, is overhead, and the lawyer should not charge for its cost "absent a contrary disclosure to the client in advance." A third-party service that bills the lawyer per use for reviewing thousands of contracts for one client is different: "it would ordinarily be reasonable for the lawyer to bill the client as an expense for the actual out-of-pocket expense incurred for using that tool." Absent an agreement, a firm's own in-house tool can be charged at no more than its direct cost plus a reasonable allocation of directly related expenses.

The Texas opinion quotes D.C. Legal Ethics Opinion 388 for the principle that under hourly billing "the economies associated with the result must inure to the benefit of the client." Clients can tell when a four-hour task is billed at the old rate, which is how this turns into a fee dispute. If you are rethinking how AI changes what you charge and what you pay for tools, our breakdown of legal research costs and what firms actually pay is a practical companion.

State Ethics Opinions That Followed 512

512 is national guidance, not a binding rule in any single state. Your state bar's opinion and your state's rules of professional conduct are what govern you. Florida's own opinion notes that "advisory ethics opinions are not binding." Counts of how many states have weighed in vary by tracker and by what counts as an opinion (formal opinions, task force reports, FAQs). We did not verify a current total in our October 2026 review, so check a current tracker or your own bar's website.

The opinions we read agree that the lawyer is responsible for the output. They differ on when you must tell the client or get consent, so read your own state's opinion rather than assuming it matches 512.

If your state appears here, read its opinion in full before relying on this summary, because the language that governs you is theirs, not the ABA's. Whether a formal opinion binds you at all is its own question, which we cover in are ABA formal opinions binding.

The Pattern: 512 Pushes the Burden Onto Your Workflow

Step back and look at the rules together.

512 does not give you a verification threshold. It does not list approved tools. It does not catalogue every situation that needs disclosure or consent; it says so directly. Over and over, the answer depends on the tool and the task.

That is the ABA being open that the technology moves too fast for bright lines. The compliance burden lands on the tool you pick and the process you build around it. Two lawyers reading the same opinion can have very different exposure depending on whether their tool learns from inputs, whether it shows sources you can open, and whether the firm has a clear policy.

The adoption numbers show where lawyers are. The ABA's 2024 Legal Technology Survey Report (released March 2025) was based on responses from 512 attorneys in private practice. In a LawSites summary of the report, 30% of respondents said they use AI-based tools, up from 11% in 2023. Among firms with 100 or more attorneys it was 46%, and among solo practitioners 18%. Accuracy was the top concern, named by 75% of respondents. And 13% said AI is already mainstream in legal practice, while 45% expect it to become mainstream within three years. These figures come from the LawSites summary, not from the report itself.

Accuracy is the concern lawyers name first. The fee and confidentiality duties tend to get less attention and carry their own risk.

A Working Checklist (That 512 Will Not Give You)

512 does not hand you a checklist, so here is an operational version built from the opinion's duties. It is practical advice built on 512, not a quotation of it.

• Pick a tool you can check efficiently. Prefer AI that shows the source it relied on, so confirming a citation takes less time than redoing the research. 512 says the amount of review depends on the tool and the task, and that tested, purpose-built tools may need less. Statutory citations count too: check U.S. Code, CFR and state code citations against primary text rather than a remembered cite.
• Know your data posture before you type. Read the tool's terms of use and privacy policy. If it is a self-learning tool and you will input client information, get informed consent first. Boilerplate in the engagement letter is not enough.
• Set firm policy and train people. Under 5.1 and 5.3, managerial lawyers must establish clear policies on permissible use, and supervisors must make sure lawyers and staff are trained. Putting the policy in writing is the practical way to do it.
• Review every authority before filing. Open it. Read the passage. Confirm it says what you claim. Check the court's local rules for AI disclosure requirements.
• Fix your billing. Do not bill the learning curve. On hourly matters, bill actual time. On flat fees, ask whether the fee is still reasonable. Before charging for an AI tool, explain the basis for the charge, preferably in writing.
• Communicate when it matters. Tell the client if they ask or require it, if you need consent, if it bears on the fee, or if the output will influence a significant decision.

FAQ

What is ABA Formal Opinion 512?

ABA Formal Opinion 512 is the American Bar Association's first formal ethics opinion on generative AI, issued July 29, 2024 by the Standing Committee on Ethics and Professional Responsibility. It applies existing Model Rules of Professional Conduct to a lawyer's use of generative AI tools, covering competence, confidentiality, communication, meritorious claims and candor, supervision, and fees.

When was ABA Formal Opinion 512 issued?

The ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, titled "Generative Artificial Intelligence Tools," on July 29, 2024. The ABA's press release of that date described it as the Association's first formal opinion covering the use of generative AI.

Which Model Rules does Opinion 512 address?

Competence (Rule 1.1), confidentiality (Rule 1.6, with Rules 1.9(c) and 1.18(b) extending similar protection to former and prospective clients), communication (Rule 1.4), meritorious claims and candor (Rules 3.1, 3.3 and 8.4(c)), supervisory responsibilities (Rules 5.1 and 5.3), and reasonable fees (Rule 1.5). The opinion also notes that other rules, such as 1.7, 1.9 and 7.1, may be implicated.

Do lawyers have to tell clients they used AI?

Not always. 512 says "depending on the circumstances, client disclosure may be unnecessary." But lawyers must disclose if the client asks, if the engagement agreement or the client's outside counsel guidelines require it, if informed consent is needed to input client information, if the use is relevant to the basis or reasonableness of the fee, or if the AI output will influence a significant decision in the representation.

Does Opinion 512 require informed consent before using AI?

In a specific situation, yes. Because many self-learning tools are designed so that their output could lead to disclosure of information, 512 says a client's informed consent is required before inputting information relating to the representation into such a tool. It also says consent is not needed when the lawyer is not inputting client information, for example when using a tool only for idea generation. General boilerplate in an engagement letter is not sufficient.

Can lawyers bill clients for time spent using AI?

Yes, for time actually spent. On hourly matters 512 says lawyers must bill their actual time, and it says a lawyer can charge for the time spent entering information into a tool and reviewing the output. A lawyer may not charge a client to learn a tool the lawyer will regularly use. On flat fees, the Rule 1.5(a) reasonableness factors still apply, and charging the same flat fee for work done much faster may be unreasonable. Before charging a client for AI tools, the lawyer must explain the basis for the charge, preferably in writing.

Is ABA Formal Opinion 512 binding on lawyers?

No. 512 says it "offers general guidance." What binds you is your own state's rules of professional conduct, and your state bar may have its own opinion on AI, which can differ from 512 on points such as consent. We cover the binding question in detail in are ABA formal opinions binding.

Does Opinion 512 require lawyers to verify AI output?

Yes, to an appropriate degree. 512 says a lawyer's reliance on or submission of AI output without an appropriate degree of independent verification or review "could violate the duty to provide competent representation as required by Model Rule 1.1." For court filings, duties to the tribunal "likewise require lawyers, before submitting materials to a court, to review these outputs, including analysis and citations to authority." The opinion does not set a fixed standard. The right amount of review depends on the tool and the task.

For the US statutes, regulations and court rules these laws sit in, see the US primary law API.

Arshita Anand

Co-Founder & CEO · Attorney

Arshita leads product and strategy at Vaquill AI, the US primary law data API. As an attorney, she decides which law we carry, how each section is sourced and cited, and what legal AI and compliance teams need from the data.