ABA Formal Opinion 512: The AI Ethics Rules for Lawyers, Explained (2026)

By The Professions Cloud Editors Updated October 13, 2026 · 5 sources · 3 min read

Prices and facts checked October 13, 2026. We don't accept payment for rankings.

Short answer: ABA Formal Opinion 512 (July 29, 2024) doesn't ban AI. It applies the Model Rules you already follow. You must understand the tool, verify everything it produces, get informed client consent before confidential information goes into a self-learning tool, supervise staff who use it, and bill fairly for time it saves. It's the ABA's first formal opinion on generative AI and still the national baseline in 2026.

The rules Opinion 512 applies

The opinion covers Model Rules 1.1 (competence), 1.4 (communication), 1.5 (fees), 1.6, 1.9(c) and 1.18(b) (confidentiality), 3.1, 3.3 and 8.4(c) (meritorious claims and candor) and 5.1 and 5.3 (supervision) (Frankfurt Kurnit summary).

1. Competence: know the tool, check the output

You need a reasonable understanding of what your AI tool can and can't do, through your own study or by consulting someone who knows. Relying on output without appropriate verification can breach your duty of competence. How much checking is needed depends on the task: brainstorming needs little, while anything filed with a court needs a full review for errors of fact or law, missing controlling authority and misleading arguments.

The stakes are real. A researcher's database now lists 2,184 decisions involving AI-hallucinated content, 1,492 in the U.S. (as of October 10, 2026) (AI Hallucination Cases Database). Recent sanctions include fines, mandatory AI training and two-year bars from practicing before a federal district court (AFP via Citizen Digital).

2. Confidentiality: informed consent for self-learning tools

This is the part that catches most firms. If a tool is self-learning (it may reuse what you input in answers for others), you need the client's informed consent before entering their confidential information. And boilerplate in an engagement letter isn't enough: you have to explain the specific risks.

You also have to vet the vendor's reliability, security and confidentiality protections, including whether they'll notify you of a breach. In practice, prefer tools that contractually don't train on your data. Some legal AI vendors state this explicitly. Spellbook, for example, advertises zero data retention agreements (Spellbook), and Paxton lists SOC 2 and ISO 27001 certifications (Paxton). Ask for the underlying reports rather than relying on badges.

3. Communication: when you must tell the client

You must disclose your AI use when the client asks, when your engagement agreement requires it, or when AI output will significantly influence a decision in the matter. Beyond that, it depends on the client's expectations and how sensitive the information is.

4. Candor: you own every filing

Rules 3.1, 3.3 and 8.4(c) mean fabricated cases or misquoted holdings are your misrepresentation, not the software's. Check every citation and quotation against the source before filing.

5. Supervision: firm policy and training

Managing lawyers should set clear firm policies on which tools can be used and how. Supervising lawyers must make sure associates and staff are trained on the tools' benefits and risks. That includes non-lawyer staff under Rule 5.3.

6. Fees: bill for time spent, not time saved

• Hourly billing: charge only for time actually spent, even if AI made you faster.
• Flat fees: a fee that doesn't reflect large AI efficiencies may be unreasonable.
• Learning time: generally not billable, unless the client asked for a specific tool you didn't know.
• Tool costs: AI built into everyday software is overhead. Per-use third-party charges can generally be passed through at actual cost.

A one-page AI policy for a small firm

1. Approved tools list, with which ones may receive client confidential information
2. Vendor check for each tool: training on data? retention? breach notice? security report?
3. Consent language for self-learning tools, explained to the client rather than buried in boilerplate
4. Verification rule: no AI-sourced citation or quote is filed until a person has pulled and read it
5. Billing rule: time actually spent; AI tool costs as overhead unless per-use
6. Training: every lawyer and staff member, at onboarding and annually

Choosing a tool? Compare options in our guide to AI tools for solo and small firms.

FAQ

Does Opinion 512 require client consent for all AI use?

No. Informed consent is required before inputting confidential client information into self-learning tools. Other uses depend on the facts, though disclosure is required in the situations listed above.

Is an engagement letter clause enough consent?

According to the opinion, boilerplate isn't enough for informed consent. You need to explain the specific risks of the tool.

Can I charge a client for the time AI saved me?

No. For hourly work, you bill the time you actually spent.

Changelog: Oct 13, 2026: first published.