What ABA Formal Opinion 512 Actually Means for Small Law Firms

What the opinion is, and what it isn’t

On July 29, 2024, the American Bar Association’s Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, titled “Generative Artificial Intelligence Tools.”⁠^[1] It was the committee’s first formal opinion on the use of generative AI in the practice of law.⁠^[2] It runs fifteen pages, and its opening sentence sets out the whole agenda:

To ensure clients are protected, lawyers using generative artificial intelligence tools must fully consider their applicable ethical obligations, including their duties to provide competent legal representation, to protect client information, to communicate with clients, to supervise their employees and agents, to advance only meritorious claims and contentions, to ensure candor toward the tribunal, and to charge reasonable fees.

Source [1]

Two things are worth knowing before reading anything into it. First, it is an advisory opinion interpreting the ABA Model Rules of Professional Conduct, as amended through August 2023.⁠^[1] A New York lawyer is bound by New York’s own Rules of Professional Conduct, not the Model Rules, so treat 512 as a careful national reference point, not as New York law. Second, the committee calls generative AI “a rapidly moving target,” and it expects that it and state and local bar ethics committees will offer updated guidance as the tools develop.⁠^[1] It even ties its central confidentiality conclusion to “the risks and capabilities of GAI tools as of the publication of this opinion.”⁠^[1]

What it does not do: it does not ban any tool, it does not require you to use AI, and it does not certify any product as safe. It tells you what you are still responsible for when you do use one.

Competence: you don’t need to be an expert, but you can’t skip the checking

Under Rule 1.1, the opinion says lawyers “need not become GAI experts,” but must have “a reasonable understanding of the capabilities and limitations of the specific GAI technology” they might use, either by learning it themselves or by drawing on someone who knows it.⁠^[1] Because these tools can produce fluent, plausible answers with no basis in fact, relying on their output without an appropriate degree of independent review could violate the duty of competence.⁠^[1] The opinion cites a Stanford study that found leading legal research companies’ generative AI systems “hallucinate between 17% and 33% of the time.”⁠^[1]

How much checking is enough depends on the tool and the task. The opinion’s own example: if you have already tested a tool by manually reviewing a smaller set of contracts and comparing its summaries against them, you would not necessarily need to re-read every contract in a large set.⁠^[1] What you can’t do is hand the tool the parts of the job that call for professional judgment, like giving advice or negotiating a claim. In the opinion’s words, “regardless of the level of review the lawyer selects, the lawyer is fully responsible for the work on behalf of the client.”⁠^[1]

Confidentiality: the part that matters most for a small firm

This is the section most small firms are actually asking about, because it decides which tools you can put client files into at all.

Rule 1.6 requires reasonable efforts to prevent unauthorized disclosure of, or access to, information relating to a representation. The opinion says that before lawyers input that information into a generative AI tool, they must evaluate the risk that it will be disclosed to or accessed by people outside the firm, and also by people inside the firm who shouldn’t see it.⁠^[1] That analysis is “fact-driven” and depends on “the client, the matter, the task, and the GAI tool used to perform it.”⁠^[1]

The “self-learning” problem

The opinion singles out tools that learn from what users put into them. With a self-learning tool, information from one client’s matter can later surface in answers to someone else’s prompts: another firm using the same product, or a lawyer in your own firm on the other side of an ethical wall.⁠^[1] The committee says the risk exists “even if the tool is used exclusively by lawyers at the same firm.”⁠^[1] Its conclusion:

Because many of today’s self-learning GAI tools are designed so that their output could lead directly or indirectly to the disclosure of information relating to the representation of a client, a client’s informed consent is required prior to inputting information relating to the representation into such a GAI tool.

Source [1]

And informed consent means something specific. The client needs your best judgment about why the tool is being used, the extent of the risk, the kinds of information involved, the ways others might use it against the client’s interests, and a clear explanation of the tool’s benefits to the matter. The opinion is direct about the shortcut many firms reached for: “merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient.”⁠^[1]

Consent is not needed when no client information goes in. Using a tool for idea generation, without inputting anything relating to the representation, is the opinion’s example.⁠^[1]

Read the terms, or have someone read them for you

As a baseline, the opinion says all lawyers should read and understand the terms of use, privacy policy, and related contract terms of any generative AI tool they use, to learn who has access to what they put in, or consult a colleague or outside expert who has.⁠^[1] Drawing on earlier opinions about cloud computing and outsourcing, it suggests that lawyers:

• make sure the tool is configured to preserve confidentiality and security, that the obligation is enforceable, and that they will be notified of a breach or of legal process seeking client information;
• look into its reliability, security measures, and policies, including limits on the provider’s liability;
• find out whether it keeps what they submit, before and after they stop using the service, and whether the provider claims rights to that information; and
• understand that the provider’s servers can fail and can be an attractive target for cyberattacks.⁠^[1]

Those obligations reach the outside company that provides the tool, too. The opinion applies Rule 5.3’s supervision of nonlawyers to generative AI providers, along with the familiar vendor diligence: reference checks and credentials, the vendor’s security policies, confidentiality agreements, and more.⁠^[1]

Telling clients, and telling courts

Separately from consent under Rule 1.6, Rule 1.4 may require you to tell clients about your AI use. The opinion says you must disclose it if a client asks, if the engagement agreement or the client’s outside counsel guidelines require it, when it is relevant to the basis or reasonableness of your fee, and when the tool’s output will influence a significant decision in the representation, such as evaluating likely litigation outcomes or jury selection.⁠^[1] Beyond that, the facts decide, and the opinion suggests the engagement agreement as a logical place to explain how you use these tools.⁠^[1]

For litigators, Rules 3.1, 3.3, and 8.4(c) mean every citation and statement of law in anything filed has to be checked. The opinion lists problems that have already arisen: citations to nonexistent opinions, inaccurate analysis of authority, and misleading arguments.⁠^[1]

If you practice in New York courts

New York has since acted on this directly. By Administrative Order AO/75/2026, the Chief Administrative Judge added Part 161, “Use of Artificial Intelligence Technology,” to the Rules of the Chief Administrator, effective June 1, 2026, for all courts of the Unified Court System, in civil and criminal cases.⁠^[3] Part 161 says the use of AI tools in preparing court papers “should not be prohibited,” and that attorneys and parties “should not be required” to disclose that use when filing, because the duties that apply to every submission already apply.⁠^[3] Individual courts may adopt their own part rules, and Part 161 offers a model rule: anyone who uses an AI tool to prepare a paper is “required to carefully review the paper and independently ensure that it contains no fabricated or fictitious cases, statutes, or other material,” and signing the paper certifies that the review was done.⁠^[3] The New York State Bar Association has published a summary of the rule for its members.⁠^[4]

Supervision: a written policy is now the expectation

Under Rules 5.1 and 5.3, the opinion says managerial lawyers “must establish clear policies regarding the law firm’s permissible use of GAI,” and supervisors must make reasonable efforts to see that lawyers and staff follow them.⁠^[1] That includes training on the basics of the technology, its limits, the ethics issues, and secure handling of client data. One practical suggestion from a footnote: mark AI-generated material as such when it is stored in a client or firm file, so later readers understand it may be fallible.⁠^[1]

For a small firm, this is less work than it sounds. It means writing down which tools are approved, what may and may not be put into each, and who checks the output, then making sure everyone, including the front desk, knows it.

Fees: bill the time you actually spent

The fee section is where the opinion gets most concrete.⁠^[1]

• Hourly work. Bill actual time. If you use a tool to draft a pleading and spend 15 minutes putting in the relevant information, you may bill those 15 minutes plus your time reviewing the draft for accuracy and completeness.
• Flat fees. Rule 1.5’s reasonableness factors still apply. If a tool lets you finish much more quickly, charging the same flat fee as without it may be unreasonable.
• Costs. A tool that works like part of equipping your office, such as an AI grammar checker in your word processor, is overhead, not a client expense. A third-party service that charges per use to review thousands of contracts for one client can ordinarily be billed as an expense at actual cost.
• In-house tools. For a tool the firm built or runs itself, absent an agreement with the client, you may charge no more than its direct cost (if any) plus a reasonable allocation of expenses directly associated with providing it.
• Learning time. You may not charge a client to learn a tool you will use regularly for clients. The exception is a client who explicitly asks you to use a specific tool you don’t know; even then, agree on the billing terms first.

Where local, on-premise AI fits

We build private AI assistants that run on hardware a practice owns or controls, so we have an obvious interest here. That is a reason to be precise about what running AI locally does and doesn’t change under Opinion 512.

The opinion’s confidentiality analysis turns on where client information goes and who can get to it afterwards. A model running on a machine in your office, not shared with other firms and not retraining itself on what your staff type into it, takes away the scenario the opinion worries about most: a matter’s details resurfacing for a later user of the same product somewhere else. The terms-of-use homework also gets simpler, because the system holding the data is yours, not a vendor’s server whose policies can change.

But a private system does not make the rest of the opinion go away:

• Competence still means checking the output. A local model can be wrong, too.
• The risk inside the firm is real. The opinion specifically warns about information crossing an ethical wall within the same firm, so any assistant that can search your files should be set up to respect who is allowed to see which matters. Ask any vendor how that is handled, us included.
• You still need the written policy and the training.
• Whoever installs and supports the system is an outside provider. The vendor diligence the opinion describes applies to that company the same way it applies to a cloud service.
• The fee rules apply wherever the model runs.

Put simply: running AI on hardware you control is one strong answer to the confidentiality question. It is not an answer to the whole opinion, and anyone who tells you it is should make you more careful, not less. If you want to see how we think about the rest, our approach to where client data goes and our straight answers to common questions are on the homepage.

A short checklist for a small firm

• List every AI tool anyone in the firm uses, including AI features built into software you already pay for.
• For each, find out from its terms whether it keeps or learns from what you put in, and who can access it.
• Decide which tools may receive client information. For any self-learning tool, plan on informed consent the opinion would recognize, not an engagement-letter clause.
• Write a short policy: approved tools, what may go into each, who reviews output, and how AI-generated drafts are marked.
• Check every citation in anything filed. In New York courts, read the part rules of the judge you are in front of.
• Review your billing practices against the fee section before the next invoice cycle.