ABA Formal Opinion 512: What It Says on AI, Confidentiality, Client Consent and Fees for Patent Work

ABA Formal Opinion 512, “Generative Artificial Intelligence Tools,” is the American Bar Association’s formal ethics opinion on lawyers’ use of generative AI, issued by its Standing Committee on Ethics and Professional Responsibility on 29 July 2024. Over 15 pages, the opinion applies the ABA Model Rules of Professional Conduct on competence (Rule 1.1), confidentiality (1.6), communication (1.4), meritorious claims and candor (3.1, 3.3 and 8.4(c)), supervision (5.1 and 5.3) and fees (1.5). It does not mention patents; the sections below apply its stated framework to invention disclosures and unpublished applications. More on AI and patent practice is on the Tech & AI hub.

Is ABA Formal Opinion 512 binding?

Not by itself. The ABA says its Model Rules “serve as models for the ethics rules of most jurisdictions” (Model Rules page), and that the committee issues opinions “to guide lawyers, courts and the public in interpreting and applying ABA model ethics rules” (ABA release, 29 July 2024). Each jurisdiction adopts its own rules (ABA adoption list). The Florida Bar’s AI opinion states: “Advisory ethics opinions are not binding.”

The opinion rests on the Model Rules as amended through August 2023 and anticipates further guidance as tools develop (p. 2 and n. 4). The ABA’s Model Rules page lists August 2026 House of Delegates amendments to the comments to several rules the opinion discusses, among them Rules 1.4, 1.5, 1.6 and 1.9; the summary below describes the opinion as issued.

Can lawyers use generative AI consistently with the duty of confidentiality?

The opinion treats it as possible, but requires a risk assessment before any client information goes into a tool. Rule 1.6 covers all information relating to the representation, whatever its source, unless the client gives informed consent, disclosure is impliedly authorized, or an exception applies; Rules 1.9(c) and 1.18(b) extend similar protection to former and prospective clients (p. 6). Rule 1.6(c) adds a duty to make reasonable efforts against inadvertent or unauthorized disclosure or access.

The opinion’s factors are the likelihood of disclosure and unauthorized access, the sensitivity of the information, the difficulty of safeguards and their effect on the representation (p. 6). Lawyers are to weigh the risk of disclosure both to people outside the firm and to people inside it who may not protect the information (p. 6). As a baseline, lawyers should read and understand a tool’s terms of use and privacy policy, or consult someone who has (p. 7).

When does the opinion require client consent to use AI?

When a lawyer will put information relating to the representation into a self-learning tool. The opinion explains that such a tool may later reveal one client’s information to users outside the firm, or to lawyers inside it who are behind an ethical wall (pp. 6–7). It concludes that “a client’s informed consent is required prior to inputting information relating to the representation into such a GAI tool” (p. 7).

Consent must be informed. The client is to have the lawyer’s judgment on why the tool is being used, the extent of the risk, the kinds of information to be disclosed, how others might use it against the client and the tool’s benefits (p. 7). General boilerplate in an engagement letter is not sufficient (p. 7). Consent is not needed where no information relating to the representation goes in, as with idea generation (p. 7). Footnote 34 ties the conclusion to the risks and capabilities of tools at publication.

How does that framework read against invention disclosures and unpublished applications?

The opinion does not discuss patent work. An inventor’s disclosure, draft claims and an application not yet published would fall within information relating to the representation as the opinion reads Rule 1.6, which covers such information regardless of source (p. 6). The sensitivity of the information is one of the opinion’s listed factors (p. 6).

On the opinion’s analysis, a practitioner who would enter that material into a self-learning tool would first need the client’s informed consent, and one who uses a tool for idea generation without entering it would not (p. 7). Where an inventor consults a firm before any engagement, Rule 1.18(b) bars the lawyer from using or revealing information learned from the prospective client, except as Rule 1.9 would permit for a former client.

The opinion does not address whether entering an invention into a generative AI tool has any effect on patentability, and no conclusion on that question is drawn here.

When must a lawyer tell a client about AI use?

The opinion says the facts of each case decide whether Rule 1.4 requires disclosure, and that disclosure may be unnecessary in some circumstances (p. 8). Rule 1.4(a)(2) requires a lawyer to reasonably consult the client about the means by which the client’s objectives are to be accomplished.

The opinion identifies cases where disclosure is required: when the client asks how the work was done or whether AI was used; when the engagement agreement or the client’s outside counsel guidelines require it; when the lawyer proposes to input information relating to the representation; when AI use bears on the basis or reasonableness of the fee; and when a tool’s output will influence a significant decision (pp. 8–9). Even where no rule requires it, the opinion calls the engagement agreement a logical place to explain AI use and record client instructions (p. 9).

What does competence require when using generative AI?

A reasonable understanding of the capabilities and limitations of the specific tool, not expertise (pp. 2–3). Rule 1.1 and its Comment [8] call for understanding the benefits and risks of the technology used, and the opinion describes the duty as continuing (pp. 2–3). Uncritical reliance on output without an appropriate degree of independent verification could violate Rule 1.1 (pp. 3–4).

The degree of review depends on the tool and the task; the opinion’s example is a lawyer who has tested a tool’s contract summaries against a manually reviewed sample (p. 4). Output may serve as a starting draft, but tasks that call for professional judgment may not be left to the tool, and the lawyer remains fully responsible for the work (p. 4). The opinion does not require lawyers to use AI, though it calls that conceivable for certain tasks in future; it says they should know the tools relevant to their work so they can decide (p. 5).

How can lawyers bill for AI time under the ABA opinion?

By charging hourly clients only for time actually spent, and by treating tool costs according to what the tool is. Under Rule 1.5, the lawyer must explain the basis of a charge for GAI tools or services before making it, preferably in writing (p. 12). Hourly lawyers “must bill for their actual time”: one who spends 15 minutes inputting information to generate a draft pleading may bill those 15 minutes and the time spent reviewing the draft (p. 12). Charging the same flat fee with and without a tool that makes the work much faster may be unreasonable (p. 12).

On costs, the opinion applies ABA Formal Opinion 93-379. A tool that works like equipping a practice, such as a grammar checker in word-processing software, is overhead, not billed absent advance disclosure; a third-party service charged per use for one client’s matter may ordinarily be billed at actual cost (p. 13). A lawyer may not bill for learning a tool used regularly for clients; where a client asks for a specific tool the lawyer does not know, billing for that learning may be appropriate if agreed first (p. 14).

What does the opinion say about candor to courts?

That AI output must be reviewed before anything is filed. The opinion cites Rule 3.1 on meritorious claims, Rule 3.3 on false statements to a tribunal and Rule 8.4(c) on dishonesty, fraud, deceit or misrepresentation, and states: “Even an unintentional misstatement to a court can involve a misrepresentation under Rule 8.4(c).” (Pp. 9–10.)

Lawyers are to review outputs, including citations, and correct errors before submitting them to a court; some courts require disclosure of AI use, and footnote 47 points to local rules (p. 10).

What must supervising lawyers do?

Under Rules 5.1 and 5.3, managerial lawyers must set clear policies on permissible GAI use, and supervisors must make reasonable efforts to see that lawyers and nonlawyers comply, including through training (p. 10). Footnote 52 adds one training suggestion: marking GAI-produced material when it is stored in a client or firm file.

For outside GAI providers, the opinion draws on earlier cloud-computing and outsourcing opinions: checking a tool’s confidentiality configuration, reliability, security and limits on liability, and whether it retains submitted information or claims rights in it (p. 11).

What have state bars said separately?

The opinion cites state and local opinions from Florida, Pennsylvania, West Virginia, the District of Columbia and California, among others (nn. 4, 32, 34 and 46). The Florida Bar’s Ethics Opinion 24-1, dated 19 January 2024, recommends obtaining the client’s informed consent before using a third-party generative AI program if that would involve disclosing confidential information.

The State Bar of California’s November 2023 practical guidance has been replaced by a revised version approved by its Board of Trustees on 14 May 2026, which says an attorney must not input confidential client information into a generative AI solution that may present material risks to confidentiality or security, absent informed client consent (p. 5).

What does the opinion leave open?

It names issues it does not resolve, including advertising under Rule 7.1 and conflicts under Rules 1.7 and 1.9, and says lawyers should consider all rules before using GAI tools (n. 4). Related reading: the explainer on AI inventorship and the site’s editorial standards.