ABA Formal Opinion 512 Checklist for Solo & Small Firms

In July 2024, the ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, its first comprehensive guidance on lawyers using generative AI (GAI). The opinion is not a rulebook so much as a mapping exercise: it walks through the Model Rules a lawyer already knows — competence, confidentiality, communication, fees, supervision, candor — and explains how each applies when a lawyer runs a matter through a large language model.

For solo and small-firm lawyers, the opinion is both a relief and a problem. A relief, because it does not ban anything or invent a new duty. A problem, because it assumes an infrastructure — vendor diligence teams, CIOs, formal training programs — that most small firms do not have. This article turns Opinion 512 into a practical checklist you can actually work through in an afternoon.

What Opinion 512 Actually Says

Opinion 512 addresses generative AI specifically (tools that produce new text, images, or code), not every form of legal technology. Its core observations are worth stating plainly:

• Competence (Model Rule 1.1). Lawyers do not need to become AI engineers, but they must understand a tool’s benefits and risks at a level sufficient to use it responsibly. That includes the risk of hallucinated citations and the limits of the tool’s training data.
• Confidentiality (Model Rule 1.6). Inputting client information into a GAI tool is a disclosure. Lawyers must evaluate how the tool stores, uses, and potentially trains on that information, and in many cases must obtain informed client consent before using it.
• Communication (Model Rule 1.4). Depending on the circumstances, clients may need to be told that AI is being used on their matter, particularly where the tool influences strategy or output the client will rely on.
• Fees (Model Rules 1.5). Lawyers cannot bill hours they did not work. If AI compresses a four-hour task into twenty minutes, the bill reflects twenty minutes. Vendor costs can be passed through only if they are reasonable and disclosed.
• Supervision (Model Rules 5.1 and 5.3). Managing lawyers must have policies in place for how the firm uses GAI, including training and audit. Non-lawyer assistance rules extend to the AI vendor itself.
• Candor and meritorious claims (Model Rules 3.1 and 3.3). Lawyers remain responsible for every citation and factual assertion. Hallucinated cases filed in court are the lawyer’s problem, not the tool’s.

Opinion 512 also flags something small firms often miss: the standard changes as the technology and its risks change. A defensible workflow in 2024 may not be defensible in 2027, and the duty of competence includes staying current.

The Checklist, Organized by Stage of the Matter

The checklist below is organized the way a matter actually moves: before you take the case, when you engage the client, while you work, when you bill, and when you close out. Each item cites the Model Rule or opinion paragraph it maps to.

1. Before You Pick a Tool: The Diligence Step Most Small Firms Skip

Opinion 512 devotes considerable attention to what it calls “reasonable understanding” of a GAI tool. For a solo lawyer, this does not mean reading a research paper. It means answering a small number of concrete questions and writing the answers down.

• Where does my input go? To a server, or does it stay on my device?
• Is my input used to train the model? Can I opt out? Is opt-out the default?
• How long is my input retained? Who at the vendor can see it?
• Are there subprocessors? Where are they located?
• Will the vendor sign a DPA or BAA if my practice requires one?

The reason to write the answers down is that Model Rule 5.3 and Opinion 512 both treat the vendor as a form of non-lawyer assistance. Supervision requires a record, however brief. A one-page memo per tool is enough for most solos.

This is also where the architecture of a tool starts to matter more than its marketing. A tool that processes content on your device cannot leak what it never receives. A tool that promises “we don’t train on your data” is only as good as the contract and the vendor’s security posture. Both can be acceptable; they carry different diligence burdens.

2. Confidentiality and the Third-Party Doctrine Problem

Opinion 512 warns lawyers that inputting client information into a GAI tool can constitute disclosure under Model Rule 1.6. In February 2026, that concern moved from theory to caselaw. In U.S. v. Heppner, Judge Rakoff in the Southern District of New York held that a litigant’s chats with a public AI platform were not privileged, drawing an explicit analogy to the third-party doctrine. The takeaway for lawyers is direct: if a consumer chatbot is a “third party” for a pro se litigant’s own communications, a lawyer who feeds a client’s privileged information into the same category of tool is inviting the same argument.

The New York City Bar reinforced the point in Formal Opinion 2025-6, issued in December 2025, which took a more restrictive line on cloud-based GAI use than Opinion 512 and emphasized informed client consent.

The practical rule that emerges from Opinion 512, NYC Bar 2025-6, and Heppner together:

Treat any GAI tool that transmits client information to a third party as a disclosure requiring analysis under Model Rule 1.6, and prefer architectures that avoid the disclosure entirely.

3. The Engagement Letter Language Question

Opinion 512 does not require a specific engagement-letter clause, but it does treat AI use as the kind of thing clients may reasonably want to know about. For most small firms, a short, plain-English clause is the safest posture. Consider three variants:

• Neutral disclosure. “The firm may use generative AI tools to assist with research, drafting, and administrative tasks. The firm reviews all AI-assisted output and remains responsible for the work product.”
• Confidentiality-focused. Adds: “The firm does not input client-identifying or privileged information into consumer AI services. Tools used in matter work are evaluated for confidentiality and, where applicable, are subject to written data-processing agreements.”
• Opt-out. Adds: “Client may instruct the firm in writing not to use generative AI tools on this matter. The firm will confirm receipt and adjust its workflow accordingly.”

For matters involving trade secrets, sealed materials, or highly regulated data, the opt-out version is usually the right default.

4. Billing: The Rule That Trips Up New Adopters

Opinion 512 is unambiguous that lawyers cannot bill hours they did not work. If AI drafts a memo in twenty minutes that would have taken four hours, the bill is for twenty minutes plus any review and revision time. Two practical consequences follow:

• Fixed-fee and value-based billing become more attractive. They align pricing with client benefit rather than hours saved, and they sidestep the awkwardness of billing dramatically less time for the same output.
• Vendor costs are pass-through, not markup. Opinion 512, consistent with earlier opinions on expenses, treats AI subscription fees as overhead absent a specific, disclosed arrangement with the client.

5. Supervision: A Firm AI Policy for Firms of One

“Firm AI policy” sounds like a document only large firms need. Model Rules 5.1 and 5.3 disagree, and Opinion 512 is explicit that even solo practitioners have supervisory duties when they use GAI. A workable solo policy fits on one page and answers five questions:

1. Which tools are approved for what tasks?
2. What categories of information may never be entered into which tools?
3. How is AI-assisted output verified before it goes to a client or a court?
4. How are AI-related costs handled on client bills?
5. Who reviews this policy, and how often?

The last item matters because Opinion 512 explicitly ties competence to keeping up with the technology. An annual review, dated and signed, satisfies the letter and spirit of the rule.

6. The Verification Step Nobody Wants to Do

The most publicized AI-ethics failures — hallucinated cases filed in court — are candor and competence failures, not confidentiality failures. Opinion 512 addresses them directly and treats them as ordinary Rule 1.1 and Rule 3.3 problems. The checklist item is simple: every citation, quotation, and factual assertion produced with AI assistance is verified against a primary source before it leaves the office.

Verification is cheaper than sanctions. Free tools like CourtListener and Google Scholar catch most fabrications; paid research services catch the rest.

7. State and Local Overlays

Opinion 512 is a floor, not a ceiling. State and local bars have layered additional guidance on top of it. The NYC Bar’s Formal Opinion 2025-6 is the most detailed to date and is required reading for lawyers licensed in New York. California, Florida, and several other states have issued their own opinions that vary in emphasis. The checklist item here is procedural: identify every jurisdiction in which you are licensed, locate the current guidance, and re-run the checklist against it.

8. Meeting Notes, Transcripts, and the Recording Problem

Opinion 512 does not address meeting recording specifically, but the confidentiality analysis extends naturally. Recorded client meetings, deposition prep sessions, and settlement calls are among the most sensitive material a lawyer handles. A cloud transcription vendor that stores audio on its servers is a Rule 1.6 disclosure to that vendor and, in the wrong architecture, potentially to other clients of the vendor.

The Brewer v. Otter.ai class action, filed in 2025, illustrates the exposure: allegations that a transcription service recorded private conversations without adequate consent turned a productivity tool into a litigation event. In West Technology Group v. Sundstrom (D. Conn. 2024), the court analyzed similar consent and recording issues in the vendor-tool context. For an internal Basil discussion of these dynamics, see On-Device vs Cloud Transcription for Lawyers.

9. The One-Page Version

If you keep only one page of this article, keep this one. Before your next client meeting:

• Know which AI tools you use, and know their data flow.
• Do not paste privileged content into consumer chatbots.
• Update your engagement letter to disclose or offer opt-out.
• Bill time worked, not time saved.
• Verify every AI-produced citation against a primary source.
• Write a one-page firm AI policy and date it.
• Re-read Opinion 512 and your state’s guidance once a year.

How Basil approaches this

Basil is a meeting notetaker built for lawyers by a practicing lawyer. Audio, transcription, and summaries are processed entirely on your Mac using the Apple Neural Engine. Nothing is uploaded, there is no server, and there are no subprocessors — Basil never receives your data. That architecture is not a promise; it is a design constraint, and it maps directly to the Rule 1.6 analysis Opinion 512 asks lawyers to perform.

Basil captures in-person meetings and virtual calls (Zoom, Teams, Meet) through on-device capture on macOS — no bot joins the call, so there is no third-party attendee to disclose or subpoena. Basil signs DPAs and NDAs on request. The Basil for Law edition — adding privilege attestation, a consent log, matter organization, and Privileged & Confidential labeling — arrives in August 2026 at $19.99/month or $199.99/year, with a 3-day trial on the monthly plan and a 7-day trial on the annual plan. The general Basil app is available today with a free tier of 60 minutes per month.

This article is for information only and is not legal advice.

Frequently asked questions

Is ABA Formal Opinion 512 binding on lawyers?

No. ABA formal opinions are advisory and interpret the ABA Model Rules. They become binding only when a jurisdiction adopts the underlying rule and the reasoning, or when a court or disciplinary body relies on the opinion. That said, Opinion 512 is treated as persuasive authority in most U.S. jurisdictions and is a reasonable baseline for a defensible AI workflow.

Does Opinion 512 require me to tell clients I use AI?

Not in every case. The opinion ties disclosure to Model Rule 1.4 and to the circumstances of the matter. Where AI meaningfully shapes strategy or output, or where the client would reasonably want to know, disclosure is expected. Many small firms handle this with a short engagement-letter clause covering all matters, which is simpler than case-by-case analysis.

Can I paste client information into ChatGPT if I turn off training?

Turning off training reduces one risk but does not eliminate the disclosure. The information still leaves your device and is retained by the vendor for some period. Opinion 512 and NYC Bar Formal Opinion 2025-6 both treat this as a Rule 1.6 disclosure that must be analyzed, and the Heppner ruling in the Southern District of New York in 2026 illustrates how courts may view public AI platforms under a third-party analysis.

How should I bill for AI-assisted work?

Bill time actually worked, including review and verification time. Do not bill the hours the task would have taken without AI. AI subscription costs are generally overhead unless you have a specific, disclosed pass-through arrangement with the client. Fixed-fee and value-based billing align well with AI-assisted workflows.

Do I really need a written AI policy if I am a solo?

Yes. Model Rules 5.1 and 5.3, and Opinion 512, treat supervision as applying even in a firm of one. A one-page policy that identifies approved tools, prohibited inputs, verification steps, billing treatment, and an annual review date satisfies the substance of the duty.

How often should I revisit my AI compliance checklist?

At least annually, and whenever you adopt a new tool or a new jurisdiction issues guidance. Opinion 512 explicitly ties competence to keeping up with the technology, so a dated annual review is both good practice and evidence of compliance.

Keep client conversations on your device

Basil transcribes and summarizes entirely on-device — no cloud, no bot, no server to subpoena. See Basil for Law → · Legal-tool reviews →

This article is for information only and is not legal advice.