ABA Formal Opinion 512 explained: what the ABA says about lawyers using generative AI

ABA Formal Opinion 512 says lawyers may use generative AI tools as long as they meet the duties they already have. It works through six of them: competence, confidentiality, communication with clients, candor and meritorious claims, supervision, and fees. For a small firm, the practical result is a short list: know your tools, check their output, get informed consent before putting client data into self-learning tools, write a firm policy, and bill honestly.

What is Formal Opinion 512?

It is a 15-page ethics opinion from the American Bar Association’s Standing Committee on Ethics and Professional Responsibility, dated July 29, 2024. It interprets the ABA Model Rules of Professional Conduct as amended through August 2023 (footnote 4).

The opinion describes itself as “general guidance” and applies the ABA Model Rules, not any one state’s rules (page 2 and footnote 4). Check your own state’s rules of professional conduct and bar opinions too. The opinion itself expects that the committee and state and local bar ethics committees will issue updated guidance as tools change (page 2).

It calls generative AI “GAI” and describes text tools as prediction tools that generate “a statistically probable output when prompted.” It also flags that some tools are described as “self-learning,” meaning “they will learn from themselves as they cull more data” (page 1). That distinction drives the confidentiality analysis.

The opinion at a glance

Duty	Model Rule	What Opinion 512 says
Competence	1.1	Understand the specific tool’s capabilities and limits. Verify output to a degree that fits the tool and task.
Confidentiality	1.6	Assess disclosure risk before inputting client information. Get informed consent for self-learning tools.
Communication	1.4	Disclose AI use when asked, when the engagement requires it, or when it affects fees or significant decisions.
Meritorious claims and candor	3.1, 3.3, 8.4(c)	Review all output, including citations, before filing.
Supervision	5.1, 5.3	Set firm policies, train lawyers and staff, and vet vendors.
Fees	1.5	Bill actual hours. Treat tools that work like office equipment as overhead. Do not bill clients for learning a tool you use routinely.

What does competence require?

Model Rule 1.1 requires competent representation, and its comment 8 includes understanding the benefits and risks of the technology you use. The opinion says lawyers “need not become GAI experts”. They need “a reasonable understanding of the capabilities and limitations of the specific GAI technology that the lawyer might use” (pages 2 to 3). Reading about legal AI tools, attending CLE and consulting people who know the technology are the examples it gives (page 3).

The core risk it names is inaccurate output, including “hallucinations,” which it describes as plausible responses “that have no basis in fact or reality” (page 3). Relying on or submitting output without an appropriate degree of independent verification could violate Rule 1.1 (pages 3 to 4).

How much checking is enough depends on the tool and the task. The opinion’s example: if you tested a tool’s contract summaries against a smaller set you reviewed by hand and found them accurate, you would not necessarily have to review every contract in the full set manually (page 4). But AI cannot be left alone to give legal advice or negotiate claims. In the opinion’s words, “the lawyer is fully responsible for the work on behalf of the client.”

It also looks ahead. It says it is conceivable that lawyers will eventually have to use these tools to complete some tasks competently. For now, it says lawyers should know which tools are relevant to their work so they can make an informed choice about using them (page 5).

When do you need client consent?

Model Rule 1.6 requires reasonable efforts to prevent unauthorized disclosure of information relating to a representation. Before inputting that information into an AI tool, the opinion says lawyers must evaluate the risk of disclosure to people outside the firm, and to people inside it who should not see it, such as lawyers behind an ethical wall (pages 6 to 7).

For self-learning tools, its conclusion is firm. Because their output could lead to disclosure, “a client’s informed consent is required prior to inputting information relating to the representation into such a GAI tool” (page 7). A footnote ties that conclusion to the tools as of publication and says the risks may change (footnote 34).

Informed consent means explaining why you are using the tool, what client information goes in, how others might use it against the client, and the tool’s benefits. The opinion is direct that “merely adding general, boiler-plate provisions to engagement letters” is not sufficient (page 7).

As a baseline, it says all lawyers should read and understand the terms of use and privacy policy of any AI tool they use, or consult someone who has. Consent is not needed when no client information goes in, such as using a tool to generate ideas (page 7).

Do you have to tell clients you use AI?

Not always. Under Model Rule 1.4, the opinion says the facts of each case decide, and disclosure may be unnecessary (page 8). It names situations where disclosure is required:

• The client asks how the work was done or whether AI was used.
• The engagement agreement or the client’s outside counsel guidelines require it.
• You plan to input client information into the tool, which triggers the consent analysis above.
• AI use is relevant to the basis or reasonableness of your fee.
• The output will influence a significant decision, such as evaluating likely litigation outcomes or jury selection (pages 8 to 9).

Even where nothing requires it, the opinion says the engagement agreement is a logical place to describe how you use AI and to record any client instructions about it (page 9).

What about fabricated citations?

Model Rules 3.1, 3.3 and 8.4(c) apply when AI output reaches a court. The opinion warns that “[e]ven an unintentional misstatement to a court can involve a misrepresentation under Rule 8.4(c).” It lists the problems seen so far as “citations to nonexistent opinions, inaccurate analysis of authority, and use of misleading arguments” (page 10).

Before filing, lawyers are to review AI output, including analysis and citations, and correct errors and omitted controlling authority. The opinion notes some courts require lawyers to disclose AI use and tells lawyers to check local rules (page 10 and footnote 47).

These are not hypothetical risks. Our Record tracks court decisions on AI-hallucinated filings by lawyers, counted by month and state.

What must partners and supervisors do?

Under Model Rules 5.1 and 5.3, the opinion says “[m]anagerial lawyers must establish clear policies regarding the law firm’s permissible use of GAI”. Supervisors must make reasonable efforts to ensure lawyers and nonlawyers follow them (page 10).

Training is part of supervision. The opinion suggests covering the basics of the technology, the tools’ capabilities and limits, ethical issues, and secure data handling. One suggestion in a footnote is to mark AI-generated material in client and firm files so later readers know it may be fallible (page 10 and footnote 52).

For AI vendors, it applies the outsourcing and cloud-computing principles from earlier opinions. Lawyers should investigate a tool’s reliability and security, confirm confidentiality obligations are enforceable, learn whether the vendor retains data or claims rights to it, and require notice of a breach (page 11).

How should you bill for AI-assisted work?

Model Rule 1.5 requires reasonable fees and expenses, and the opinion says you must explain the basis for any AI charge before charging it, preferably in writing (page 12).

• Hourly. Lawyers “must bill for their actual time”. The opinion’s example: if a pleading takes 15 minutes of prompting, bill those 15 minutes plus the time spent reviewing the draft (page 12).
• Flat or contingent. Rule 1.5(a)‘s factors still apply. If AI makes a task much faster, charging the same flat fee as without it may be unreasonable (page 12).
• Tool costs. If a tool functions like equipping and maintaining a practice, its cost is overhead unless you told the client otherwise in advance. An AI feature built into word processing software, such as a grammar checker, should be treated as overhead. A third-party tool billed per use to review thousands of contracts for one client can ordinarily be passed through at actual cost (page 13).
• Learning time. You may not bill a client to learn a tool you will use regularly. If a client asks for a specific tool you do not know, billing for learning it may be appropriate, ideally under a written agreement (page 14).

Do state bars say the same thing?

Check your own state. Opinion 512 itself cites state and local guidance, including Florida Bar Ethics Opinion 24-1, a joint Pennsylvania and Philadelphia opinion, a West Virginia opinion and D.C. Bar Opinion 388 (footnotes 4, 34 and 46).

California is one example of how state guidance has moved since. The State Bar of California’s committee says its 2026 Practical Guidance replaces its 2023 version and adds a section on “agentic” AI, meaning tools that carry out tasks without step-by-step prompting. On confidentiality it reaches a similar result to the ABA: a lawyer must not input confidential client information into a tool that may present material risks to confidentiality or security without informed client consent. It also says lawyers must not permit AI systems to autonomously file documents or communicate with the court, a point Opinion 512 does not address.

What should a small firm do first?

The steps below are our own checklist, not the ABA’s. Each one draws on a section of the opinion covered above:

1. List the AI tools anyone at the firm uses, including AI features inside software you already pay for.
2. Read each tool’s terms of use and privacy policy. Note whether it trains on your inputs.
3. Write a one-page policy on which tools are allowed, for which tasks, with which client data.
4. Train every lawyer and staff member on that policy and on checking citations.
5. Update your engagement letter to describe AI use in plain terms, and get specific informed consent where client data goes into a self-learning tool.
6. Review your billing so hourly entries reflect actual time and AI tools that work like office equipment stay in overhead.