Are ABA Formal Opinions Binding? What Opinion 512 Means for Your AI Workflow

No, ABA formal opinions are not binding. They do not bind a court, a disciplinary board or a lawyer, until a state authority chooses to give them weight. An ABA formal opinion is the ABA ethics committee's reading of its Model Rules, and a court or state bar may follow it, cite it, or ignore it. ABA Opinion 512 is no exception.

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

A question lawyers keep asking: if Opinion 512 isn't binding, why does everyone say I have to follow it?

The short answer is that both statements can be true. Opinion 512, the ABA's July 29, 2024 opinion on generative AI, binds nobody on its own. If you want the rule-by-rule duty detail, we wrote a companion piece on what Opinion 512 actually requires.

This post is about the prior question: what kind of authority is this document, and why does "not binding" turn out to be cold comfort.

TL;DR

• ABA formal opinions are persuasive authority and carry no force of law by themselves. The ABA is a private membership organization. Its Model Rules become law in a state only when that state adopts them, and its ethics opinions interpret those Model Rules without the force of a statute or a court ruling.
• Opinion 512 creates no new duties. It applies existing Model Rules to generative AI in six areas: competence (Rule 1.1), confidentiality (Rule 1.6), communication (Rule 1.4), meritorious claims and candor to the tribunal (Rules 3.1, 3.3 and 8.4(c)), supervision (Rules 5.1 and 5.3) and fees (Rule 1.5). Your own state's version of those rules already binds you.
• It is stricter on confidentiality than many readers assume. For self-learning tools, 512 says a client's informed consent is required before you input information relating to the representation, and boilerplate in an engagement letter is not enough.
• States do not wait for the ABA. Florida issued its own AI ethics opinion about six months before 512, and California published guidance before 512 and replaced it in May 2026.
• Your state's version is the one that applies. Many state and local bars have published their own generative-AI guidance, and counts vary because trackers count different things (formal opinions, practical guidance, court rules). Look up your own bar.

Part of our legal AI verification and hallucination guide series.

For related verification, hallucination and vendor-trust coverage, see ABA Formal Opinion 512 Explained: What Every Lawyer Using GenAI Must Do and "We Do Not Train on Your Data": How to Verify the Claim.

What "Binding" Actually Means Here

Lawyers throw the word "binding" around loosely, so let us be precise, because the precision is the whole point.

The American Bar Association is a voluntary professional association. It does not license lawyers and it does not discipline them.

What it produces are Model Rules of Professional Conduct. They are a template that states can adopt, change or reject. A state's own authority, often its highest court, decides which rules govern its lawyers. California shows the process: its Supreme Court approved a new set of Rules of Professional Conduct by Administrative Order 2018-05-09, effective November 1, 2018.

That answers a common mix-up. California adopted a comprehensive new set of rules in 2018 (the State Bar's current rules page says so). They use a numbering scheme like the Model Rules, for example Rule 1.4 on communication with clients, but the text differs in places. Read the California version, not the Model Rule.

Formal opinions sit one layer further out. A formal opinion does not propose a new rule. It is the ABA's Standing Committee on Ethics and Professional Responsibility saying how it reads the existing Model Rules in a new situation. Bars say the same about their own opinions. The Florida Bar's Opinion 24-1 opens with the line "Advisory ethics opinions are not binding."

An ABA opinion is persuasive. A court or a state bar may find it convincing and cite it, the way a judge cites a respected treatise or a sister-state decision. Nobody is obligated to agree. A disciplinary board in your state can read 512, disagree with a passage, and go its own way.

So when someone tells you Opinion 512 is not binding, they are correct in the narrowest, most technically defensible sense. The mistake is what they do with that fact.

Why "Not Binding" Is the Wrong Place to Stop

Two things are true at once. First, 512 itself commands nothing. Second, everything 512 describes is already binding on you through a different door.

The opinion does not invent a verification duty or a confidentiality duty. It applies the Model Rules on competence (1.1), communication (1.4), fees (1.5), confidentiality (1.6), meritorious claims and candor to the tribunal (3.1, 3.3 and 8.4(c)) and supervision (5.1 and 5.3) to lawyers who hand work to a machine.

Those rules, in your state's adopted version, already bind you. They bound you before 512 existed. They will bind you if 512 is withdrawn tomorrow.

The lawyers in Mata v. Avianca were not sanctioned for violating an ABA opinion. On June 22, 2023, more than a year before 512, Judge Castel sanctioned two lawyers and their firm $5,000, jointly and severally, "pursuant to Rule 11, or, alternatively, its inherent authority," for submitting non-existent judicial opinions with fake quotes and citations created by ChatGPT, then standing by them after the court questioned their existence. No ethics opinion was involved. We traced that whole arc and the verification protocol that prevents it in our piece on AI hallucinations and sanctions.

Courts keep reaching the same place through existing rules. In Prososki v. Regan, 321 Neb. 38 (Mar. 20, 2026), the Nebraska Supreme Court struck an appellate brief full of fictitious authority, dismissed the appeal and referred the lawyer to the state's Counsel for Discipline, citing Nebraska's own rules on candor, competence and misconduct. Its closing caution: "Whether using AI or not, the obligations of candor, competency, diligence, and making good faith arguments remain the same."

This is why "is it binding" is the wrong question. The duties are binding. 512 is a clear map of how they apply to AI.

Persuasive authority is not nothing. It is the thing a decision-maker can cite when they rule against you.

States Do Not Wait for the ABA

If you think the ABA leads and the states follow, the timeline says otherwise.

Florida acted about six months before the ABA. The Florida Bar issued Opinion 24-1 on January 19, 2024, about six months before the ABA published 512 on July 29, 2024. California's State Bar had published its Practical Guidance on generative AI two months before Florida, in November 2023.

California has since replaced that guidance. The State Bar's current Practical Guidance says it "replaces the 2023 version" and, at the California Supreme Court's request, addresses agentic AI. The State Bar's Ethics and Technology Resources page records that its Board of Trustees approved the update on May 14, 2026. If you practice in California, cite the 2026 version. The Legislature has also acted: California's SB 574 (Chapter 858, Statutes of 2026), signed on September 30, 2026, adds Business and Professions Code § 6068.1. It requires a lawyer who uses generative AI to take reasonable steps to verify the output, including every case and statutory citation, and to disclose the use of generative AI to the court "for all documents submitted to the court." It also amends Code of Civil Procedure § 128.7 so that no paper filed in a California court may contain a citation the responsible attorney has not personally verified. The bill has no urgency clause, so it takes effect January 1, 2027 (bill text and status; Governor's September 30, 2026 signing announcement).

Colorado has put AI into its rules of professional conduct. Colorado Supreme Court Rule Change 2026(02), adopted January 8, 2026 and effective immediately, adds Scope paragraph [20A] (technology, including artificial intelligence, "does not diminish a lawyer's responsibilities under these Rules") and a new Comment [9] to Rule 1.1, which lists the other Rules AI use can implicate, among them 1.4, 1.5, 1.6, 3.1, 3.3, 5.1 to 5.3, 7.1 and 8.4(g), and adds that reliance on technology does not diminish the duty of independent judgment. It also amends Comment [8] to Rule 1.1 to cover the benefits and risks of relevant technology.

Pennsylvania and Philadelphia also went their own way. The joint formal opinion of the Pennsylvania Bar Association and the Philadelphia Bar Association (2024-200, May 22, 2024) lists "Maintaining Transparency" among its duties: lawyers "should be transparent with clients, colleagues, and the courts about the use of AI tools in legal practice." Some courts also require disclosure by standing order, and 512 tells lawyers to check the local rules of the court they are in.

That ordering shows states treat AI ethics as their own jurisdiction to regulate. They read their own rules, saw lawyers already using these tools, and issued direction.

How many bars have followed depends on what you count. Formal opinions, practical guidance, task force reports and court rules are different things, and trackers that count them differently report different totals. The Courthouse Digital tracker lists official guidance jurisdiction by jurisdiction with a link to each source, and says itself that it is not a complete survey, so confirm anything it shows against your own bar's site.

The practical takeaway: the authority you answer to is not one ABA document. It is your state's version, which may echo 512 and may raise the floor above it. Treat 512 as the national baseline and then read what your own jurisdiction did, because that is the document a disciplinary board in your state will apply directly.

What Most People Get Wrong

Two errors, equal and opposite.

The first is the one we have been picking apart: treating "not binding" as "ignorable." That reasoning works right up until a complaint lands, at which point the underlying rule (which is binding) gets enforced and 512 can be cited as the standard of reasonable care.

The second error is the inverse: treating 512 as the rule, a self-contained AI statute you can comply with by reading one PDF. It creates no new duties, it draws few bright lines (the right level of review of AI output depends on the facts), and it does not control in your state.

Lawyers who memorize 512 and stop there can still be caught by a state opinion or a court order that asks for more.

The honest framing sits between them. 512 is a floor, set out nationally, that your own state has probably built on.

Whatever the label on the document, you still own three things personally: verification of what the tool produces, confidentiality of what you feed it, and honest billing for the time it saves or costs. None of those obligations cares whether you call its source "binding."

So What Does This Mean for Your AI Workflow

Strip away the authority taxonomy and the operational reality is concrete. The compliance variable you control is the tool you pick and the workflow you wrap around it.

Verification. 512 does not set a fixed amount of checking. It says the appropriate level of review depends on the task and the tool, and that before submitting materials to a court a lawyer must review AI outputs, including analysis and citations, and correct errors. 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 by hand. We lay out the step-by-step routine in how to verify AI legal citations before filing.

Confidentiality. Here 512 is firm. Because many self-learning tools can disclose what you input, it says a client's informed consent is required before you put information relating to the representation into one, and boilerplate in an engagement letter is not enough. It also tells lawyers to read and understand the tool's terms of use and privacy policy, or have a colleague or outside expert do it. If you put nothing about the client's matter into the tool, this consent is not required. Whether a tool trains on your inputs, how long it retains them, and what its data-processing terms say are compliance questions, not just procurement ones.

Primary sources. A surprising amount of "AI compliance" is the old duty to check primary authority. When you are verifying that a statute says what a brief claims, you want the text of the statute from an official or reliable source, not a paraphrase.

Fees. The quietest duty in the 512 cluster is Rule 1.5. A lawyer may not charge a client to learn about a tool the lawyer will regularly use for clients, with an exception when a client explicitly asks for a specific tool the lawyer does not know. Efficiency gains from AI also raise honest questions about flat fees versus hourly billing. That is an engagement-letter problem, and it is the one most likely to surface in a fee dispute rather than a malpractice claim.

The Real Question to Ask Instead

Stop asking whether ABA Formal Opinion 512 is binding. It is not, and the answer leads you nowhere useful.

Ask instead: what has my state's highest court adopted, what has my state bar said about AI specifically, what do the local rules of the court I am in require, and have I picked tools and built a workflow that make verification cheap, confidentiality defensible, and billing honest?

Those are the questions a disciplinary board would be asking after the fact, with 512 open on the desk as persuasive authority. Better to answer them first.

FAQ

Are ABA formal opinions binding?

No. The American Bar Association is a private membership organization, not a regulator, so its formal opinions are persuasive authority and do not carry the force of law. An opinion matters to you when your state's highest court or bar adopts or applies the underlying rule. A court or disciplinary board can read an ABA opinion, disagree with it, and rule the other way.

What is ABA formal opinion authority?

It is persuasive, the same kind of weight a court gives a respected treatise or a sister-state decision. A formal opinion is the ABA Standing Committee on Ethics and Professional Responsibility explaining how it reads the Model Rules in a new situation. Decision-makers cite it when they find it convincing, but nobody is obligated to follow it.

Is ABA Opinion 512 binding?

No. Opinion 512, issued July 29, 2024, creates no new duties and binds no one by itself. What binds you are the duties it applies (competence, confidentiality, communication, candor, supervision and fees) as adopted in your own state's rules of professional conduct. Those rules bound you before 512 and would survive its withdrawal.

What actually binds a lawyer instead of an ABA opinion?

Your state's rules of professional conduct, as adopted and enforced by your state's highest court, plus the rules of the court you are appearing in and any court order in your case. State bar ethics opinions help interpret those rules but are often advisory too; Florida's Opinion 24-1 begins "Advisory ethics opinions are not binding." California adopted a new set of rules effective November 1, 2018, and its text differs from the Model Rules in places.

Do all states follow ABA Opinion 512?

No. States may follow it, diverge from it, or go further. Many state and local bars have published their own generative-AI guidance, and counts vary with what a tracker includes. California published guidance in 2023 and replaced it with a 2026 version, Florida issued Opinion 24-1 in January 2024, and Pennsylvania and Philadelphia issued a joint opinion that stresses transparency about AI use. Check what your own bar has published.

Does Opinion 512 require lawyers to disclose AI use to clients or courts?

Not every use. 512 says client disclosure may be unnecessary depending on the circumstances, but a lawyer must disclose if the client asks, if the engagement terms require it, when the lawyer proposes to input information relating to the representation into a self-learning tool (which requires the client's informed consent), and when AI use bears on the basis or reasonableness of the fee. For courts, 512 notes that some courts require disclosure and tells lawyers to consult the applicable court's local rules.

A closing takeaway

The label on the document is the wrong fight. The duties under it (verification, confidentiality, supervision, fees) are the work. Pick tools whose data flow you can defend in writing, and a workflow that makes cite-checking cheap enough that you actually do it.

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.