What ABA Opinion 512 means for AI in estate planning tech

On July 29, 2024, the American Bar Association’s Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, "Generative Artificial Intelligence Tools." The opinion runs fifteen pages. It maps six existing Model Rule obligations onto lawyer use of generative AI: competence, confidentiality, communication, candor, supervision, and fees. The opinion permits lawyer AI use under strict conditions. Its more consequential move: treating AI as a nonlawyer assistant under Model Rule 5.3 and making the supervising attorney personally responsible for what the AI produces. For advisors evaluating legal-tech platforms that rely on attorney-reviewed templates, Opinion 512 is the document to read. The supervision duties run downstream through every platform built on that model.

The six duties ABA 512 maps to generative AI

Opinion 512 clarifies how six existing Model Rules apply when a lawyer uses a generative AI tool in practice, rather than inventing a new rule set. The timing matters. The opinion followed two years of public sanctions in cases where lawyers filed briefs containing hallucinated citations generated by consumer AI tools, including widely reported sanctions in federal court in 2023 that surfaced fabricated case law cited as authority. The ABA had to address the pattern before it hardened into an ethics crisis. Opinion 512 is the result.

• Competence (Model Rule 1.1): a lawyer must understand the AI tool’s capabilities and limitations well enough to use it responsibly. Blind deployment is malpractice under the existing competence duty.
• Confidentiality (Model Rules 1.6, 1.9(c), 1.18(b)): the lawyer must evaluate the risk that client information is disclosed to or accessed by third parties. For self-learning or public-API AI products, informed client consent may be required before client information is entered.
• Communication (Model Rule 1.4): the lawyer may need to disclose AI use to the client, especially when the client asks or when the AI’s output materially affects the representation.
• Candor (Model Rules 3.1, 3.3, 8.4(c)): the lawyer has a duty to verify AI outputs. The well-publicized hallucinated-case-citation sanctions in 2023 and 2024 are addressed here explicitly.
• Supervision (Model Rules 5.1, 5.3): managerial lawyers must establish firm policies on AI use. Generative AI used inside a firm is treated analogously to a nonlawyer assistant and requires supervision under Model Rule 5.3.
• Fees (Model Rule 1.5): lawyers can only bill for time actually spent. The AI tool’s cost is generally not separately billable as a disbursement without client disclosure.

The Model Rule 5.3 hook

Opinion 512’s extension of Model Rule 5.3 to generative AI is the single most important provision for anyone building or buying legal technology. Model Rule 5.3 governs lawyer responsibility for nonlawyer assistance. Historically that meant paralegals, contract staff, outsourced document review, and technology service providers. Opinion 512 pulls AI tools inside the same framework.

Under Model Rule 5.3(a), managerial lawyers in a firm must establish policies reasonably assuring the nonlawyer’s conduct is compatible with the lawyer’s professional obligations. Under 5.3(b), any lawyer with direct supervisory authority over the nonlawyer has the same obligation. Under 5.3(c), a lawyer is responsible for a nonlawyer’s conduct that would violate the Model Rules if engaged in by the lawyer. Opinion 512 applies each of those subsections to generative AI.

The practical effect: the supervising attorney cannot outsource AI-output review to the AI itself. The lawyer cannot delegate the supervision function to a commercial AI vendor’s disclaimer language. The lawyer must build, document, and enforce a human review layer over every AI-assisted output that reaches a client.

In practice, supervision under Opinion 512 looks like written policies on approved AI tools, documented training for the lawyers and staff who use them, a review protocol that flags AI-generated content for human verification before filing or client delivery, and incident logging when AI output is found to contain errors. None of this is optional under the opinion. A bar complaint reviewer looking at a lawyer’s AI-use practices is going to look for the documentation. Absence of documentation is itself evidence.

Where the state bars landed

State bar opinions issued in the same cycle mostly aligned with Opinion 512’s framework, with a few distinctive emphases. Florida Bar Op. 24-1, issued January 19, 2024, was the first state bar opinion in the country on lawyer AI use and emphasized confidentiality, oversight under the state analog to Model Rule 5.3, billing discipline, and restrictions on chatbot marketing. Several other jurisdictions followed with opinions across 2024 and into 2025, including a New York City Bar formal opinion distinguishing closed-system AI from public-API AI, a District of Columbia Bar opinion articulating a two-question confidentiality test under Rules 1.1, 1.5, and 1.6, a Kentucky Bar opinion clarifying that routine AI-assisted research does not require disclosure absent special circumstances, and a Texas opinion covering competence, confidentiality, verification duties, and fair billing.

These opinions vary in detail. They converge on a single message. The attorney is responsible. The AI is assistive. Supervision must be real. The primary-source links for each opinion are in this journal’s sources index.

A few distinctions worth flagging for legal-tech buyers. Opinions from jurisdictions with stricter confidentiality regimes tend to require closer scrutiny of the AI tool’s terms of service, especially clauses that allow the vendor to train on submitted inputs. Opinions addressing billing tend to require fee reductions when AI materially shortens attorney time on a matter, which matters for platforms that market "AI-assisted document preparation" to law firms as a productivity tool. Several opinions distinguish between closed-system AI (deployed inside a firm, no external data exposure) and public-API AI (sending inputs to a third-party service), with sharply different confidentiality consequences. A platform that runs inputs through a public API without informed client consent is on thin ice under those opinions.

The supervision inheritance chain

Opinion 512 binds lawyers. Legal-tech platforms themselves sit outside its direct scope. The opinion still creates a second-order effect that matters for every document preparation platform on the market, and the category has not named it yet. Call it the supervision inheritance chain.

Every legitimate legal-tech platform in the estate planning category sells "attorney-reviewed templates" as part of its value proposition. The value of that review is directly downstream of the reviewing attorney’s own professional compliance. If the reviewing attorney used generative AI in the review, Opinion 512 governs how that use was supposed to be supervised. If the attorney failed to supervise the AI under Model Rule 5.3, or failed to verify output under the candor duty, or failed to establish competence in the tool under Rule 1.1, the review itself is only as good as the supervision that backed it.

The inheritance runs in a straight line:

• A platform sells attorney-reviewed templates to advisors as part of a subscription.
• An advisor subscribes, relies on the attorney review, and offers estate planning as a service to clients.
• A client completes the questionnaire and signs a template-generated document.
• If the template later produces a bad outcome, litigation traces back to the review.
• The reviewing attorney’s Opinion 512 compliance comes into play at the review step.
• If the attorney’s review was not 512-compliant, the chain breaks at that link and the value of the platform’s "attorney-reviewed" claim collapses with it.

Advisors buying legal tech sit three steps away from Opinion 512. They are still downstream of it. The question worth asking is not whether the platform uses AI. The question is whether the reviewing attorneys comply with Opinion 512 when they review.

A concrete scenario makes the chain visible. An advisor using a legal-tech platform finalizes an estate plan for a household. A template clause in the trust document, which the platform marketed as "attorney-reviewed," later produces an outcome that contradicts the grantor’s stated intent at signing. The household sues. Discovery surfaces that the template was reviewed by an attorney who used a consumer-grade AI tool to summarize the clause before signing off. The attorney did not verify the AI summary. Under Opinion 512, that is a competence and candor failure. The attorney’s professional liability carrier now has a coverage question, the platform’s vendor has a marketing-claim question, and the advisor’s firm has a reliance question. Everyone downstream of the original review has a problem.

Opinion 512 diligence questions for legal-tech buyers

If you are evaluating a legal-tech platform for estate planning in 2026, the Opinion 512 diligence questions below test whether the supervision inheritance chain holds. A vendor’s answers tell you more than any marketing page.

Platforms that answer each question clearly are on the right side of the supervision chain. Platforms that respond with vague marketing language are offering a review that may not survive the Opinion 512 framework if it ever gets tested in a bar complaint or a malpractice suit.

A good vendor answer sounds like: "Our templates are reviewed by Attorney X, admitted in State Y (bar number Z), and the review is documented with date, scope, and methodology. The reviewer does not use public-API AI tools during the review. Template updates are re-reviewed before they ship." A bad vendor answer sounds like: "We have a team of legal experts who review all our content with the latest technology." The first answer passes Opinion 512 diligence. The second answer is marketing copy that collapses on any scrutiny. If you are evaluating platforms, pick the vendors who can give you the first answer in writing. Our competitive comparison walks through the same evaluation rubric applied to the estate planning platforms advisors typically consider.

How Bancroft handles attorney review

Bancroft’s template library is reviewed by licensed attorneys in the jurisdiction each document applies to. The review covers clause text, the questionnaire logic that assembles clauses from client answers, the statutory citations each clause relies on, and the execution pages. The review also runs through generated output across a set of representative client scenarios, so the reviewing attorney examines the final document rather than only the raw clause text.

Reviewed files are locked. Every reviewed template file has a SHA-256 cryptographic hash recorded in a manifest checked into the codebase. A continuous integration test runs on every build and recomputes the hashes from disk. Any change to a reviewed file breaks the build until an attorney re-reviews the change and the manifest is updated. Shipping a clause edit without a new attorney sign-off is not possible through the normal deployment path. The lock mechanism is a product constraint enforced in the CI pipeline.

When a household’s specific situation exceeds what the templates cover, the platform flags the document for attorney review at $299. Lady Bird Deeds require mandatory attorney review at $399 because the tax and titling consequences are sensitive enough that no template should be executed without an attorney confirming the specific facts. Amendments and restatements to existing Bancroft documents are always free and move through the same human attorney review path. Each of these is a separate engagement from the template-level review described above. Advisors who want to trace the supervision chain for a specific document type can ask for the reviewing attorney’s name and methodology as part of the demo process.

The deeper product mechanics for the attorney review flow are covered in our essay on attorney-reviewed vs attorney-prepared. The audit-logging and access-control infrastructure that supports individualized attorney review is described on the security page.

The bottom line

Opinion 512 is the clearest map the ABA has drawn of how existing Model Rules apply to generative AI. The opinion permits lawyer AI use and requires real supervision of how the AI is used. The lawyers who do not supervise their AI end up on the wrong side of a bar complaint, and every legal-tech platform that relies on those lawyers for template review inherits the problem.

For advisors evaluating platforms in 2026, the question is simpler than the AI hype cycle suggests. Ask who reviewed the templates. Ask how those reviewers supervised their tools. Ask what happens when a template changes. The platforms whose answers survive those three questions are the ones whose attorney-reviewed claim still means something.

The rest are selling a word.