What should a law firm AI use policy include

ABA Formal Opinion 512 (July 29, 2024) is the ABA Standing Committee's guidance on generative artificial intelligence tools. It does not create new rules. It maps existing Model Rules onto that use: competence (1.1), communication (1.4), fees (1.5), confidentiality (1.6), candor (3.3), and supervision (5.1 and 5.3).

California attorneys should also keep the California State Bar Practical Guidance on the Use of Generative Artificial Intelligence in the Practice of Law (November 2023) next to Opinion 512. The Practical Guidance is California State Bar ethics guidance for generative AI in practice. It is not a Bar endorsement of any course, vendor, or template. A national template is a starting point. It is not a finished California policy until a lawyer admitted in California adapts it.

Enforceable checklist vs template-only framing

Template pages answer "give me pasteable clauses." This page answers "what must the written policy actually cover so a partner can supervise." Both matter. A template without the seven buckets is hard to enforce. Seven buckets without jurisdiction-specific adaptation is incomplete for California.

An enforceable policy is the document a managing partner can point to when an associate asks which tool is approved, when a client asks what data left the firm, or when a court asks who checked the citations. If the answer is only "we told everyone to use AI carefully," the firm does not yet have a supervision instrument—it has a reminder.

• Map each bucket to a named Model Rule (and to the California Practical Guidance where it speaks to the same duty).
• Name approved tools and banned paste destinations in the same document supervisors will use.
• Put the vendor diligence questions in writing before a pilot has client files in it. See questions to ask an AI vendor before a law firm signs.
• Treat verification as a filing gate, not a personal preference.
• Write the incident path before you need it. See what to do if a hallucinated citation already reached a filing.

Scannable checklist: the seven buckets

Use this list as a coverage check. If a bucket is missing, blank, or only a slogan, the policy is not yet supervisable.

1. Tool — Named products and allowed tasks are written down.
2. Client information — What may and may not be pasted is written down (Rule 1.6).
3. Client communication — When the client is told is written down (Rule 1.4).
4. Review — Who checks citations and propositions before filing is written down (Rules 1.1, 3.3).
5. Supervision — How partners oversee lawyers and staff is written down (Rules 5.1, 5.3).
6. Fees — What may be billed for tool use is written down (Rule 1.5).
7. Incident — What happens after a fake or unverifiable citation is filed is written down.

The seven buckets to write down

1. The tool. Which products are approved, and for which tasks. A policy that says "use AI carefully" does not name a system a supervisor can enforce. Separate research assistance, drafting assistance, and anything that touches discovery or client uploads. If a consumer chatbot is banned for client matter work, say so in the same place the firm names the approved stack.
2. Client information. What may be pasted into a consumer chatbot, and what may not. Rule 1.6 is the confidentiality duty. Opinion 512 and the California Practical Guidance both treat confidentiality as a fact-specific risk analysis, not a slogan. Informed consent, when it is required, is not a one-line boilerplate in an engagement letter. Write which categories of matter data may leave the firm network, which must stay in a contracted tool with written terms, and who may authorize an exception.
3. Client communication. When the lawyer tells the client a tool is being used, including when the client asks, when the engagement requires it, and when the use is material to a significant decision. Rule 1.4. The policy should say who documents that conversation and where the note lives in the file.
4. Review before it leaves the office. Who checks citations, quotes, and legal propositions before a filing or a client letter. Uncritical reliance can implicate Rule 1.1 and Rule 3.3. Require a human verification step against a reporter, docket, or other primary source before generative output is treated as authority. "I asked the tool again" is not verification.
5. Supervision. What partners and managers do when associates, paralegals, or contract attorneys use a tool. Rules 5.1 and 5.3. Include a training step, not only a prohibition. Name who owns the policy, how often it is reviewed, and what happens when someone bypasses the approved tool list.
6. Fees. Bill for time actually spent directing and reviewing the tool. Do not bill the client to learn a tool the firm will use generally. Rule 1.5. If the firm charges a technology or research fee that includes AI tooling, say so in engagement terms the client can understand—not in a buried footer.
7. The incident. What happens if a citation that cannot be verified has already been filed. Stop treating it as authority, inventory related unverified material, notify supervising counsel, correct the tribunal as that court's rules require, tell the client if material, keep the record, and harden the policy. See what to do if a hallucinated citation already reached a filing.

California overlay: Practical Guidance next to Opinion 512

For California practices, read the November 2023 Practical Guidance as a parallel map, not a substitute. Opinion 512 is national Model Rules guidance. The Practical Guidance is California State Bar guidance on generative AI in the practice of law. Neither document is a finished firm policy. Neither is an endorsement of alcbar.com or of any vendor.

In practice, the California overlay usually tightens three drafting choices: (1) confidentiality and vendor terms before client information is entered; (2) competence and verification before generative output leaves the office; and (3) supervision so associates and staff are not left to invent their own tool rules. If your policy cites only Opinion 512 and never names the Practical Guidance, a California supervisor still has a gap to close. For CLE that maps both documents, see California MCLE that covers ABA Formal Opinion 512.

How this differs from a free template download

Pasteable clauses help once you know what they must cover. They do not replace naming tools, naming paste bans, naming who signs citations, and naming the incident path. Download a template if it saves drafting time. Then run the seven-bucket checklist against it. If the template is silent on supervision training, fees, or the already-filed hallucination path, fill those gaps before you call the document done.

Vendor diligence belongs upstream of the policy's "approved tool" list. A tool that looks convenient in a demo can still be the wrong place for client files if training-on-inputs, subprocessor access, retention, and deletion are only dashboard toggles. Put those questions in writing first: AI vendor questions before a law firm signs.

FAQ

What should a law firm AI use policy include?

A usable firm AI policy covers seven buckets: the approved tool, what client information may be entered, when the client must be told, how output is reviewed before filing, supervision of lawyers and staff, how fees are billed, and what happens if a citation that cannot be verified has already been filed. Ground the policy in ABA Formal Opinion 512 and, for California attorneys, the State Bar Practical Guidance (November 2023).

Does ABA Formal Opinion 512 require a written firm AI policy?

Opinion 512 does not hand you a finished policy. It maps generative-AI use onto existing Model Rules and tells the lawyers who manage the firm which duties have to be written down so supervisors can enforce them.

What California guidance should a firm AI policy also reflect?

California attorneys should also read the California State Bar Practical Guidance on the Use of Generative Artificial Intelligence in the Practice of Law (November 2023) alongside Opinion 512. A national template is a starting point, not a finished California policy.

How is an enforceable firm AI policy different from a free template download?

A template answers "give me pasteable clauses." An enforceable policy answers "what must be written so a partner can supervise." Name approved tools, banned paste destinations, who signs off on citations, how fees are billed, and the incident path. Map each bucket to a Model Rule and to California Practical Guidance where it speaks to the same duty.

A template is a starting point. It is not a finished policy until a lawyer admitted in that jurisdiction adapts it. The on-demand course covers what to write down under Opinion 512 and the California Practical Guidance. Related: vendor questions, hallucinated citation already filed, and CA MCLE & Opinion 512.

The on-demand California MCLE course is $249. It includes the five modules, the 111-page reading, and the 35-question assessment. California MCLE accreditation is pending. The certificate is issued when the provider number is assigned. The course does not guarantee compliance, prevent sanctions, or satisfy a court, insurer, or client.

Enroll Now — $249

Course detail.