ABA Formal Opinion 512 in Practice: What It Means at Your Desk

On July 29, 2024, the ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512 — its first formal ethics opinion on generative AI. The press coverage called it guidance; that undersells it. The opinion takes six Model Rules and applies them to AI use in a way that gives every lawyer a concrete standard: competence, confidentiality, communication, supervision, candor, and reasonable fees. This article is the desk-level translation — what each requirement actually means when you are choosing a tool, drafting an engagement letter, or reviewing a filing.

It is the operational companion to the rest of this series — the privilege analysis and the citation-verification workflow both trace back to duties this opinion states.

First, what the opinion is — and is not

Opinion 512 is advisory. The ABA Model Rules are not the rules of professional conduct in any jurisdiction until a state supreme court adopts them, and the opinion says so. But two facts make it the de facto national baseline. First, 49 of 50 states (California is the outlier) follow the Model Rules' core structure, so the opinion's rule-by-rule framing maps cleanly onto most state rules of professional conduct. Second, state bar AI opinions issued since July 2024 routinely cite and track its analysis, layering state-specific requirements on top rather than departing wholesale. The practical consequence: if your AI practices cannot be documented against the six duties below, a malpractice carrier, disciplinary investigator, or sanctioning court already has a ready-made framework for identifying the gaps.

The opinion's own framing of the whole duty set is worth quoting in full:

"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."

Rule 1.1 — Competence: understand the tool, don't master it

The competence standard is more achievable than most lawyers fear. You do not need to become a GAI expert. What the opinion requires:

"To competently use a GAI tool in a client representation, lawyers need not become GAI experts. Rather, lawyers must have a reasonable understanding of the capabilities and limitations of the specific GAI technology that the lawyer might use."

Three practical points follow. First, the duty is ongoing — the opinion calls technological competence "not a static undertaking," so a one-time training does not satisfy Rule 1.1 years later if the tool, the model, or your use has materially changed. Second, the required verification is task-dependent: "there is no single answer," the opinion explains — document review requires far more independent review than using a tool to generate ideas. Third, the stakes are stated plainly: "lawyers' uncritical reliance on content created by a GAI tool" is not just risky; it is "almost certainly malpractice." Reading about legal AI, attending CLE, and consulting colleagues or experts are the approved ways to stay current — and keeping training records is the documentation that proves you did.

Rule 1.6 — Confidentiality: the informed-consent threshold

The confidentiality analysis starts with the broad duty: all information relating to the representation is protected. Applied to AI, the opinion sets a clear threshold based on what kind of tool you are using:

"[B]ecause 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."

Two things matter here. First, the trigger is self-learning tools — tools that learn from what you submit. If the tool is not self-learning, or you are not inputting client information (idea generation without client data), consent is not required. Second, boilerplate will not do:

"To obtain informed consent when using a GAI tool, merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient."

Informed consent means explaining why the tool is being used, the specific risks, the kinds of client information that will be disclosed, how others might use that information, and the benefits to the representation. And before any of that, the opinion requires a baseline vendor review: read the Terms of Use, privacy policy, and related contractual terms to answer four questions — who has access to inputs and outputs, whether and how the provider retains data, whether the tool trains on submitted content, and what happens if access is revoked. That review can be delegated, but it must happen, and it must be documented per tool.

Rule 1.4 — Communication: when disclosure to the client is required

The opinion imposes no blanket duty to disclose every use of AI. Disclosure is required in specific situations: if the client asks; if the engagement agreement or outside counsel guidelines require it; when informed consent is required under Rule 1.6; when AI use is relevant to the basis or reasonableness of the fee under Rule 1.5; and when the tool's output will influence a significant decision in the representation. The opinion notes the engagement letter is the logical place to make disclosures — and many firms now add a plain-language paragraph naming the categories of tools they use and inviting the client to ask for detail.

Rule 1.5 — Fees: bill actual time, don't charge for learning

The fee analysis has two direct operational consequences. First, efficiency is not billable at the old rate:

"GAI tools may provide lawyers with a faster and more efficient way to render legal services to their clients, but lawyers who bill clients an hourly rate for time spent on a matter must bill for their actual time."

Second, learning time is not the client's to pay for:

"[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."

The opinion also distinguishes tool costs that are overhead (a grammar-check bundled into word processing) from client-specific pass-through expenses (a per-matter charge from a contract-review vendor) — and firms should map each approved tool to one category and disclose the billing treatment in advance. If a flat fee was set on pre-AI time assumptions, the opinion signals it may no longer be reasonable once AI materially compresses the work.

Rules 5.1 and 5.3 — Supervision: policy, training, and vendor diligence

The supervision duties have both an internal and an external face. Internally:

"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."

That means a written firm AI policy naming approved tools, prohibited uses, and supervisory responsibility; training records for every attorney and staff member who touches an approved tool; and a supervision protocol for AI-assisted work product before it leaves the firm. Externally, Rule 5.3(b)'s reasonable-efforts duty applies to the AI provider itself, carrying forward the diligence framework from the ABA's cloud-computing and outsourcing opinions: reference checks and credentials, security policies, confidentiality agreements, conflicts screening, and whether the provider retains or claims proprietary rights to submitted content. That vendor-diligence file is the documentation a disciplinary investigation would ask to see.

Rule 3.3 — Candor to the tribunal: the sanctions line

The candor analysis is where the opinion meets the sanctions cases. On the risk, the opinion is blunt:

"Even an unintentional misstatement to a court can involve a misrepresentation under Rule 8.4(c). Therefore, output from a GAI tool must be carefully reviewed to ensure that the assertions made to the court are not false."

And the pre-filing review duty is broader than checking citations:

"In judicial proceedings, duties to the tribunal likewise require lawyers, before submitting materials to a court, to review these outputs, including analysis and citations to authority, and to correct errors, including misstatements of law and fact, a failure to include controlling legal authority, and misleading arguments."

The courts have been enforcing exactly this line. Beyond Mata v. Avianca's $5,000 sanction (S.D.N.Y. 2023), the cases have escalated: in Wadsworth v. Walmart, 348 F.R.D. 489 (D. Wyo. 2025), the court revoked the lead attorney's pro hac vice admission and fined him $3,000, with two co-counsel fined $1,000 each for motions citing nonexistent cases. In Jason M. Hatfield, P.A. v. Pirani (W.D. Ark. 2025), where the defendant admitted using AI to draft post-trial motions citing nonexistent cases and quotations, the court imposed attorney-fee and cost awards exceeding $1.5 million — the largest AI-citation sanction on record. The line between a $5,000 reminder and a seven-figure award is drawn by the same Rule 3.3 duty, applied with increasing severity.

The documentation that proves compliance

The opinion does not require a compliance bureaucracy — it requires evidence that the duties were met. Concretely, the documentation a carrier or disciplinary body is likely to ask for:

• A written AI policy (Rules 5.1, 5.3) — dated, naming approved tools, prohibited uses, and supervisory responsibility.
• A vendor-diligence file per approved tool (Rules 1.1, 1.6) — Terms of Use, privacy policy, retention, training-on-input behavior, and the four-question review answered.
• Tool-specific informed-consent language (Rule 1.6) — meeting the "not boilerplate" standard, for any self-learning tool receiving client information.
• Training records (Rules 1.1, 5.1, 5.3) — every user trained on each approved tool, logged by date, refreshed on material change.
• A pre-filing verification log (Rule 3.3) — AI-assisted filings logged, with the citation-verification step documented by the reviewing attorney.
• A billing policy for AI time (Rule 1.5) — hourly rules, flat-fee treatment, overhead versus pass-through classification per tool.
• An incident-response procedure (Rule 1.6) — what happens if client information is exposed, including notification.

None of this is exotic; it is the ordinary evidence of competent, supervised practice applied to a new tool. The firms that keep these records do not fear the opinion — they use it as the specification for how to run AI practice defensibly.

What Lawyer Assistant does about this

The opinion's duties map onto Lawyer Assistant's design in a direct way. The confidentiality duty (Rule 1.6) is served by architecture: everything runs locally on your machine, so the four-question vendor review has a short answer — there is no vendor with access, no retention, no training on your data. The competence duty (Rule 1.1) is served by transparency: every answer cites the exact source text it was grounded in, so the verification the opinion demands is a click rather than a search. The candor duty (Rule 3.3) is served by the same design — traceable citations make the pre-filing review the opinion requires straightforward. The policy, training, and billing documentation remain yours, as they always are. The tool handles the architecture; the opinion's duties belong to the lawyer.

The bottom line

Opinion 512 is not a warning — it is a specification. It tells you, rule by rule, what responsible AI use looks like: understand the tool, protect the data, disclose what the rules require, bill actual time, supervise everyone and everything in the chain, and verify everything before it reaches a tribunal. Every requirement maps to a concrete action and a document you can keep. The firms that treat the opinion as a checklist rather than an abstraction are the ones that will use AI aggressively and safely — and the ones that will have the records when someone asks.

Sources & further reading

• ABA Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512, "Generative Artificial Intelligence Tools" (July 29, 2024) (PDF) — the primary source; all quotations in this article are from the opinion.
• Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023) — $5,000 sanction for six fabricated ChatGPT citations.
• Wadsworth v. Walmart Inc., 348 F.R.D. 489 (D. Wyo. 2025) — pro hac vice revoked and $3,000/$1,000 fines for citing nonexistent cases.
• Jason M. Hatfield, P.A. v. Pirani (W.D. Ark. 2025) — attorney-fee and cost awards exceeding $1.5 million after AI-drafted filings cited nonexistent cases and quotations.
• North Carolina State Bar, 2024 Formal Ethics Opinion 1 (Nov. 1, 2024) — state application of the same framework, including Rule 1.6(c) and Rule 5.3.

This article is general information about technology and professional practice. It is not legal advice for any specific matter, and rules vary by jurisdiction — verify against the authority applicable to your matter.