How to Write an AI Policy for a Small Law or Accounting Firm

Your staff are probably already using AI tools, whether or not the firm has approved any. A short written policy turns scattered habits into clear rules. This guide explains what to include and gives you a template to fill in.

Why a small firm needs one

For law firms, ABA Formal Opinion 512 says lawyers with managerial authority must set clear policies on the firm’s permitted use of generative AI. Supervising lawyers must also make reasonable efforts to ensure that lawyers and staff follow them, under Model Rules 5.1 and 5.3.

For accounting firms, the AICPA Code of Professional Conduct bars members in public practice from disclosing confidential client information without the client’s specific consent. When a member uses a third-party service provider, the Code’s interpretation on disclosing information to such providers says the member should either have a confidentiality agreement with the provider and reasonable assurance about its safeguards, or get the client’s specific consent first. Next to that interpretation, the Code points readers to a nonauthoritative AICPA staff article on ethics threats when members use AI. Tax preparers have additional federal rules on using and disclosing return information, covered in a separate guide.

A policy is how a small firm shows it has thought these duties through. It also gives staff a simple answer when they are unsure.

Keep it short

A policy nobody reads protects no one. Aim for two pages. Write rules, not essays. Put the reasoning in training, not in the policy.

Use plain words. “Do not paste client names into any tool not on the approved list” works better than a paragraph on data minimization.

What to include

Approved tools

List each tool by name and plan, for example “ChatGPT Business, firm workspace” rather than “ChatGPT”. Consumer and business versions of the same product can have different data terms. For each tool, say what it is approved for.

Anything not on the list is not approved. Say that explicitly.

What may and may not be entered

Sort information into tiers. A common split:

• Public or non-client material. Published law, public filings, general research questions. Allowed in any approved tool.
• Client information. Names, facts, documents, financial data. Allowed only in tools approved for client data, and only when the matter permits it.
• Restricted information. Data covered by a protective order, a client’s outside counsel guidelines, a client’s instruction against AI use, or special rules such as tax return information or health records. Not allowed unless a named partner approves in writing.

Review requirements

Every AI output is a draft. Say who must review it and how closely. Require extra checking for anything going to a court, a regulator, a taxing authority or a client. For legal work, that means checking every citation against the source.

Client consent and disclosure

State when the firm tells clients about AI use and when it asks for consent. For law firms, Opinion 512 says informed consent is needed before entering client information into a tool that learns from inputs, and that boilerplate in an engagement letter is not enough. For accounting firms, follow the Code’s third-party provider interpretation and any client instructions.

Record-keeping

Decide what you record. Useful items include which tool was used on which matter, any client consents, and a marker on documents that started as AI drafts. Opinion 512 suggests marking AI-generated material in files so later readers know to check it.

Training

Set a minimum: everyone trained before first use, and again when tools or rules change. Cover what the tools do well, where they fail, what may not be entered, and how to report a problem.

Incident reporting

Mistakes will happen. Someone will paste the wrong document into the wrong tool. Make reporting fast and blame-free so people report early. Name who to tell, and how quickly.

For law firms, ABA Formal Opinion 483 explains lawyers’ duties after a data breach, including notifying clients when material client information is involved. Accounting firms should check state breach-notification laws and their own professional obligations. The policy should route incidents to the person who makes that call.

Fill-in template

Copy this, replace the bracketed text, and delete what does not apply.

[Firm name] AI Use Policy Effective [date]. Policy owner: [name, title]. Next review: [date].

1. Purpose. This policy sets the rules for using artificial intelligence tools in firm work. It applies to all partners, employees, contractors and interns.

2. Approved tools. Only these tools may be used for firm work:

Tool and plan	Approved for	Client information allowed?
[Tool, plan]	[Drafting, research, summaries]	[Yes / No]
[Tool, plan]	[Use]	[Yes / No]

Any tool not listed is not approved. Personal accounts may not be used for firm work. Request a new tool from [policy owner].

3. What may be entered.

• Public and non-client material: any approved tool.
• Client information: only tools marked “Yes” above, and only for the matter you are working on.
• Never enter, without written approval from [partner]: [material under protective order, tax return information, health information, government ID numbers, passwords, other categories].
• Follow any client instruction or outside counsel guideline that limits AI use.

4. Review. All AI output is a draft. The responsible [lawyer / CPA] must review it before it is used. Anything filed, sent to a client or sent to a third party must be checked line by line. Every legal citation and every figure must be checked against its source.

5. Clients. [Describe how the engagement letter explains AI use.] Before entering client information into any tool that may use inputs to train or improve its models, obtain the client’s informed, specific consent and record it in [location].

6. Records. Record [tool used, matter, date, consents] in [system]. Mark documents that began as AI drafts with [label].

7. Training. Complete firm AI training before first use and [annually / when tools change]. [Policy owner] keeps the training log.

8. Incidents. If client information is entered into an unapproved tool, or an AI error reaches a client, court or agency, tell [name] within [time]. Do not try to fix it quietly. [Name] decides whether clients or others must be notified.

9. Billing. [Law firms: Bill actual time spent. Do not bill clients for learning a routine tool. Treat general AI subscriptions as overhead unless disclosed in advance.] [Accounting firms: describe your approach.]

10. Violations. Breaking this policy may lead to [consequences].

Signed: [name], [date]. I have read and will follow this policy: [employee signature].

Putting it into practice

Start with an inventory. Ask everyone, without judgment, which AI tools they use for work today. You will likely find personal accounts and AI features inside software you already pay for.

Then check each tool’s data terms before you approve it. The due-diligence checklist in our related guide gives you the questions.

Hold a 30-minute meeting to walk through the policy. Have everyone sign it. Put a review date on the calendar.

If you want a broader framework, the National Institute of Standards and Technology publishes a voluntary AI Risk Management Framework and a profile for generative AI. Most small firms will not need it, but it is a useful reference.

How Zrico fits

Zrico gives each firm its own private workspace and databases. Claude is the default and is processed only in the US. GPT and Gemini are off unless the firm owner turns them on, and their processing is not limited to the US. No AI provider trains on prompts, files or answers. Owners and admins get a usage and audit report and retention settings. Zrico does not replace your policy or make the firm compliant, and its output must be reviewed by a professional.

Sources

• ABA Formal Opinion 512, Generative Artificial Intelligence Tools (July 29, 2024): https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf
• ABA Formal Opinion 483, Lawyers’ Obligations After an Electronic Data Breach or Cyberattack (October 17, 2018): https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-op-483.pdf
• Model Rule 5.1, Responsibilities of a Partner or Supervisory Lawyer: https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_5_1_responsibilities_of_a_partner_or_supervisory_lawyer/
• Model Rule 5.3, Responsibilities Regarding Nonlawyer Assistance: https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_5_3_responsibilities_regarding_nonlawyer_assistant/
• AICPA Code of Professional Conduct (see 1.700.001, Confidential Client Information Rule, and 1.700.040, Disclosing Information to a Third-Party Service Provider): https://pub.aicpa.org/codeofconduct/ethicsresources/et-cod.pdf
• NIST AI Risk Management Framework: https://www.nist.gov/itl/ai-risk-management-framework

General information, not legal advice. Check the current rules in your jurisdiction.