ABA Formal Opinion 512: What It Says About Lawyers Using Generative AI

ABA Formal Opinion 512 is the ABA's first formal opinion on generative AI in law. The group issued it on July 29 of 2024. The opinion walks through what Model Rules apply when a lawyer uses a generative ai tool at work.

This piece is Opinion 512 explained in plain words. It covers what the opinion says and what a working lawyer should check. The article does not give legal advice; it sums up a public text for a reader in a law firm.

What is ABA Formal Opinion 512?

The ABA issued Formal Opinion 512 on that same date. The ABA Standing Committee on Ethics and Professional Responsibility wrote it. That group puts out formal opinions on how the aba model rules fit new topics as the field changes.

Formal Opinion 512 addresses lawyers' duties when they are using generative artificial intelligence in a law firm. State bars adopt the Model Rules of Professional Conduct in their own words.

The opinion cites nine Model Rules. Those rules are 1.1, 1.6, 1.4, 3.1, 3.3, 8.4, 5.1, 5.3, and 1.5.

The American Bar Association writes formal opinions as guidance. State and local bar groups then look to that aba guidance on generative ai in their own work. They use it as they write their own bar ethics materials and ethics rules on artificial intelligence.

ABA opinion 512 summary: the six duties at a glance

The opinion names six duties and shows how each fits when lawyers are using generative ai tools at work. These are the duties the opinion walks through.

• Competence (Rule 1.1): know the tool they use, what it does, and when the ai output needs more review.
• Privacy (Rule 1.6): protect client data, and for a self-learning gai tool, get client consent first.
• Communication (Rule 1.4): tell the client when the client would want to know. Some cases call on you to disclose ai use up front.
• Candor to the court (Rules 3.1, 3.3, 8.4(c)): review every output, since false cites and false facts are still false.
• Supervision (Rules 5.1 and 5.3): managing lawyers set a firm policy, and the firm trains on it.
• Fees (Rule 1.5): a fee must still be fair, and learning time is not billable to the client.

Each duty gets its own section in the opinion. The next parts walk through each one in turn.

What the opinion says about competence with an AI tool

Model Rule 1.1 asks for competent work. That rule covers the benefits and risks of ai technology a lawyer uses.

This opinion fits that duty to lawyers who use ai tools on client work. A lawyer does not need to be a GAI expert.

The lawyer must have a sound grasp of the gai tool in use. That means knowing the main features and the main limits.

A lawyer may learn the tool by self-study, legal classes, or asking a peer. The lawyer must also keep up as the tool changes.

The opinion calls tech competence not a static task. A new feature in a common tool may look nothing like last year's.

The opinion names one well-known limit: AI hallucinations. A tool may give an answer that sounds right but is wrong. An ai feature may invent a case, change a quote, or fill in facts it never saw.

Blind use of ai output that was generated by AI is a competence problem. The fix is your own review, and the level turns on the task.

A short contract summary from a general tool may need less review. Legal advice, drafts, and court filings need more.

For a step-by-step method, see a framework to verify AI-generated legal content. The opinion also keeps the floor clear.

No AI model stands in for a lawyer's judgment. A lawyer may not leave legal advice to the tool alone.

When does a lawyer need client consent to use a GAI tool?

Model Rule 1.6 covers privacy. A lawyer must keep client data private, in and out of the firm.

The opinion fits that rule to inputs a lawyer types into a gai tool. Before any such input, weigh the risk that others may see the data.

"Others" means people outside the firm. It also means people inside the firm who should not have access.

The opinion flags one class of tool: self-learning gai tools, a form of self-learning ai. That kind keeps learning from what users type into it.

The same data may show up later in another user's output. For that class of tool, a lawyer must get client consent first.

Consent must be informed. The opinion says boiler-plate text in an engagement letter is not enough.

The client needs an honest view on why the tool is used, and what data will go in. The client also needs to know what risks come up.

With those facts in hand, the client decides. Before any such use generative ai work begins, read the Terms of Use for ethical ai use.

Ask who sees the input, whether the tool keeps it, and whether the input leaves the tool's servers. Not every use of ai needs consent.

A tool used only for idea help, with no client data going in, may not need it.

When must a lawyer tell a client about an AI tool?

Model Rule 1.4 covers contact with the client. The opinion explains when lawyers should share their ai use apart from the privacy rule.

A client who asks about using ai gets an answer. The opinion says a lawyer should share such use when the client asks how the work was done.

The same rule fits when the engagement letter or outside counsel rules call for it. A lawyer should also raise the subject when the output will drive a key choice.

One case is using an AI tool to predict a likely outcome. Another is using a tool to help with jury picks.

A client would want to know whose judgment is in play. The lawyer also has to talk about the use of generative ai when it changes the fee.

The fees section below covers that point. Say a client hired the lawyer for personal skill.

Quietly handing part of the work to a tool can clash with what that client expects. The rule does not call for sharing every time.

A tool that plays a small, routine role may need no mention.

What the opinion requires for AI in court filings

Model Rules 3.1, 3.3, and 8.4(c) tell lawyers to bring only real claims. They also bar false conduct in a court.

The opinion fits those duties to any ai output in a filing. It lists problems that already show up in court.

Those include cites to cases that do not exist, wrong readings of real cases, and false claims. Any one of these can turn into sanctions.

A full picture lives in the post on Mata and the regulatory regime that followed. The duty is easy to state.

Before a court sees any output, the lawyer reads it and checks every cite. Check every quote against the real case, and every fact against the record.

A mistake is not excused because a tool made it. Rule 8.4(c) covers false conduct even when the error is by accident.

A lawyer who signs a filing is on the hook for what is in it. State courts have added rules in the same direction.

For a plain reading of 2026 state rules, see the state court AI rules article.

How a law firm supervises its AI tools

Model Rules 5.1 and 5.3 cover oversight of others in the firm. Rule 5.1 covers managing and overseeing other lawyers.

Rule 5.3 covers nonlawyer help, which can include outside vendors. The opinion asks each law firm to put a clear policy in writing.

Who may use what tool, for which tasks, and with which data? A good law firm ai policy answers those questions and names the rules everyone must follow.

Training comes next. The firm must train each person who may use generative ai at work. The training covers the basics, the risks, and the rules.

The opinion suggests marking any work the AI made. Later readers then know how the work came about.

Rule 5.3(b) reaches outside the firm. When a firm uses an outside AI vendor, the firm must make a real effort.

The vendor must meet the firm's duties too. That means checking the vendor's safety steps and reading its contract.

The opinion ports its cloud and outsourcing checklist to AI vendors. The main items:

• Set up the tool to protect client data, and put the duty in writing.
• Confirm the vendor will tell the firm about any data breach or any legal demand for client data.
• Learn the tool's reliability, safety, and liability terms before you sign.
• Find out whether the tool keeps data after the lawyer stops using it.
• Treat vendor servers as a target for attack, and plan for that risk.

A firm that follows that list has solid proof. It met its duties for its AI program.

Can a lawyer bill a client for AI work?

Rule 1.5 of the Model Rules asks fees to be fair. The opinion walks through what that means for AI use.

A lawyer who bills by the hour must bill actual time, and that rule is not new. The opinion fits it to AI.

If a tool helps a lawyer finish a task in less time, the client pays for less time. A lawyer may not bill the hours the task would have taken with no tool.

The opinion also keeps a long-standing rule on learning. A lawyer may not bill a client for learning a tool.

The tool will be used for many clients, so the cost is overhead. One case is different.

If a client asks for a tool the lawyer does not yet know, the client and lawyer may agree. They can set the lawyer's skill time at the client's cost.

That agreement is best in writing. What about charging for the tool itself?

The opinion sorts AI costs into two kinds. Tools baked into general office software are overhead.

Grammar help inside a word processor is one such case. The lawyer may pass per-use charges from an outside AI service to the client as an actual cost.

The lawyer should share the basis in writing. Flat fees and fees paid on win must still be fair.

If a tool saves a lot of time, the same flat fee may no longer match the work.

How Opinion 512 fits with state ethics guidance

Opinion 512 is national in its posture, but each state bar still runs its own rules. State work matters too.

The opinion cites state and local opinions that reached close answers under their own rules. State bars have issued their own ethics opinions on AI at a steady pace.

Florida Bar Opinion 24-1 is one of them. Florida's committee issued it in January 2024.

It covered many of the same duties and flagged ad issues under Florida Rule 4-7.13. Pennsylvania and Philadelphia put out a joint formal ethics opinion in 2024.

California issued day-to-day guidance in 2024, and West Virginia put out its own text in 2024. Other states have added more since.

A full list of such materials changes every few months. The real result for lawyers is to check both sources.

The national text gives the baseline duties, and the state source may add state-only detail. Where a state rule is stricter, it controls for lawyers in that state.

Court rules add another layer. State court rules and federal court standing orders are a separate matter from these rules.

For related reader questions about ChatGPT in practice, see chatgpt for lawyers: what courts require.

What should a working lawyer check now?

Opinion 512 does not hand a reader a checklist. It leaves a set of habits to build.

These are the habits the opinion points toward.

• Pick a tool and learn it. Read the Terms of Use, and test the tool on sample work before you rely on it.
• Decide when client data goes in. For any self-learning tool, plan to get consent, and draft a short script you can use.
• Set a firm policy. Write down which tools are okay, for what tasks, and with what data each one may touch.
• Review every output. Verify cites and quotes against the source, and do not sign a filing unread.
• Price the work honestly. Do not bill hours the work did not take, and do not pass overhead as a cost.
• Watch state guidance.Add your state's opinion on AI to your reading list, and check for updates often.

A lawyer who builds these habits is in a better spot. A court or a client may later ask how the work was done.

The duties the opinion names do not shift much. The habits around them do not either.

A clean ai ethics habit scales with the tools as they change.

Where Opinion 512 fits: AI in legal practice

Opinion 512 is one piece of a broader landscape of ai use. More aba formal opinions may come as new questions arise.

Lawmakers and courts keep adding rules on AI regulation. State bars keep issuing their own guidance.

Common themes repeat across those sources. Each source looks at a few core challenges of AI.

There is the risk of made-up output. There is the risk of leaked client data.

There is also the risk of a lawyer leaning on a tool for a human-judgment call. Aba formal opinions and state opinion sources share that frame.

Most leave big-picture AI governance to future rules. General-purpose AI tools and legal-specific tools raise the same duties about use of gai tools.

The opinion does not carve out legal tools. The same six duties apply.

A lawyer using a chat tool for ideas has the same floor. So does a lawyer using a research tool for cites.

Expect ai adoption and ai use in legal practice to keep growing. More guidance from bar groups will follow.

For most lawyers, the big question is now day-to-day. Pick a tool, set the rules, and train the team.

Where to read Opinion 512

The opinion is public. The committee posts it as a PDF on its site.

The citation is ABA Formal Opinion 512 from 2024. It runs 15 pages, with footnotes.

A short course in legal ethics helps a reader place it inside the rest of the rulebook. The opinion uses dense legal English.

The rule cites map back to the Model Rules themselves. A reader new to the Model Rules may want to read the rule first.

Then read the section of 512 that covers it. Then read them together a second time.

That pairing is where the work on these rules really starts for a working lawyer. Rules in this area are changing.

ABA formal opinions on tech land at their own pace. State bars keep issuing guidance on these tools.

State and federal courts keep updating their rules. This article reflects Formal Opinion 512 as adopted.

It also reflects the related state opinions and court rules as of November 2026.