Law Firm AI Policy Template

What this template does

"They're asking firms, 'Do you use AI? Do you police it? Do you have protocols in place?'" Source: Stan Sterna, SVP and risk control lead at Aon, in Digital Insurance / Accounting Today, April 2026. A written AI policy answers the second and third questions. The firm's usage register answers the first.

A written AI policy is the first document a malpractice carrier typically requests at renewal when asked about AI use. It is also the first deliverable required by Rules 5.1 and 5.3 in ABA Formal Opinion 512. Built section-by-section against Opinion 512, the template fits firms of 5 to 50 attorneys without dedicated compliance staff. The resulting documentation reads as serious to a carrier underwriter, and it addresses the obligations a disciplinary investigator would expect a firm to have considered.

Every section below maps to a Model Rule and to the specific passage in Opinion 512 that the section documents. For the full rule-by-rule walkthrough, see ABA Formal Opinion 512: A Compliance Guide.

Why a policy, not a ban

A blanket ban on AI is a common first instinct and a poor second strategy. Bans drive use to personal devices and free-tier consumer tools, where the firm has no visibility into where client data goes and most providers train on submitted content by default. A documented policy that permits supervised use under defined conditions produces better client-confidentiality outcomes than a ban a firm cannot enforce.

A ban is also incomplete on the facts. AI features are already embedded in tools attorneys use every day: Westlaw, Lexis, Microsoft 365, Adobe Acrobat, and most major document-management and practice-management systems. A policy that addresses only standalone chatbots fails to govern the AI that has already arrived.

Recent state-bar guidance reflects this consensus. The North Carolina Bar Association, in January 2026 guidance, framed it directly: prohibition drives usage underground; clear policies bring it into the open where it can be supervised. The template below operationalizes that posture.

What's in the template

Fifteen sections. Each section maps to a Rule of Professional Conduct (RPC) and to the passage in Opinion 512 that drives it:

• Scope and definitions (Opinion 512, Introduction)
• Tool capabilities and limitations (RPC 1.1; Opinion 512 § A)
• Approved AI tools with risk tiers (RPC 1.1, 1.6; Opinion 512 §§ A, B)
• Acceptable use (RPC 1.1, 1.5; Opinion 512 §§ A, F)
• Prohibited uses (RPC 1.6, 8.4(g); Opinion 512 § B)
• Client data and confidentiality with data classification (RPC 1.6; Opinion 512 § B)
• Client disclosure and communication (RPC 1.4; Opinion 512 § C)
• Verification of AI-assisted filings (RPC 3.3, 8.4(c); Opinion 512 § D)
• Data security and privacy (RPC 1.6; Opinion 512 § B)
• Supervision and training (RPC 5.1, 5.3; Opinion 512 § E)
• Billing treatment (RPC 1.5; Opinion 512 § F)
• Competence and training requirements (RPC 1.1; Opinion 512 § A)
• Incident response (RPC 1.6, 1.4)
• Exceptions and waivers (RPC 5.1)
• Policy review and update cadence (RPC 1.1, 5.1)

The template

The sections below are the policy itself. Adapt bracketed placeholders ([like this]) to your firm. The language is written to be used close to as-is; trim where a section does not apply to your practice.

1. Scope and definitions

This policy governs how all attorneys and staff of [Firm Name] ("the Firm") use generative artificial intelligence ("AI") tools on firm work and client matters. "AI tool" means any software that generates text, images, audio, video, code, or other content in response to a user prompt. The term covers general-purpose tools (ChatGPT, Claude, Gemini, Copilot), legal-specific tools (Harvey, Spellbook, Westlaw Precision AI, Lexis+ AI, Paxton AI), and AI features embedded in software the Firm already uses. The policy reaches firm devices, personal devices used for firm work, and any third-party platform that processes firm or client information. Personal AI accounts, including free-tier and individually-paid consumer subscriptions registered to an attorney or staff member, are prohibited for firm or client work regardless of device.

2. Tool capabilities and limitations

AI tools assist with editing, summarizing, brainstorming, and refining written content. They are not, and should not be relied on as, authoritative legal research databases, sources of binding legal authority, or substitutes for professional judgment. AI output may be inaccurate, outdated, or biased; staff must critically assess every result before relying on it. Where a task can be done correctly only with primary-source legal research, the primary-source research is done in the authoritative database (Westlaw, Lexis, or equivalent), and AI is not a substitute. Use of an AI tool for a task does not transfer responsibility for the output away from the responsible attorney.

3. Approved AI tools

The Firm maintains a written list of approved AI tools. Only tools on the approved list may be used in connection with firm or client work. The approved-tools list is maintained by [Managing Partner / AI Committee / designated partner] and is reviewed at least annually and whenever a new tool is proposed.

Approval requires, at minimum:

1. Review of the provider's Terms of Use, privacy policy, and data-handling representations.
2. Confirmation of whether the provider trains on submitted content, and under what conditions.
3. Confirmation of whether and how data is retained after use.
4. Confirmation of the jurisdictions in which data is stored and processed.
5. Documentation of the intended firm use case.

Current approved tools: [list each tool, the approved use case, and the partner responsible for reviewing it].

Approved tools are classified into three tiers based on the data and use cases they support:

• Tier 1 (Approved). Vetted enterprise tools with no model training on submitted content, contractual confidentiality protections, and recognized security evidence (SOC 2 Type II, ISO 27001, or equivalent). May be used for any approved use case, subject to the data classification in Section 6.
• Tier 2 (Conditional). Tools approved for specific limited use cases: administrative drafting, internal brainstorming, or summarization of public information, without confidential client information. Use outside the approved use case requires Tier 1 review or matter-specific informed consent.
• Tier 3 (Prohibited). Consumer-grade and free-tier tools, personal AI accounts, and any tool that retains or trains on submitted content without an enterprise contract limiting that practice. Includes free versions of ChatGPT, Claude, Gemini, Copilot, and similar consumer products.

Diligence for each approved tool is documented against the Vendor Due Diligence Checklist and retained in the vendor file.

4. Acceptable use

Approved AI tools support the categories of work below. All use remains subject to Section 6 data classifications and Section 8 verification requirements:

• Drafting and editing. Clarity, grammar, structure, and readability improvements to internal and client-facing documents, after the responsible attorney establishes the substantive content.
• Summarizing and condensing. Summaries of internal memos, deposition transcripts, discovery materials, or other content the firm has lawful access to and has reviewed for confidentiality.
• Brainstorming and issue-spotting. Issue lists, argument outlines, alternative framings, and discussion points, to be developed and verified by an attorney.
• Research support. Candidate lines of inquiry, candidate authorities, or alternative theories. Section 8's verification protocol applies to every authority the AI tool surfaces.
• Internal training and education. Hypothetical fact patterns, practice exercises, or training materials that contain no client-identifying or privileged information.
• Administrative content. Non-client communications, internal procedures, marketing copy, and similar firm-facing content.

AI assists attorney work; it does not substitute for it. Where a task requires authoritative legal research, professional judgment, or the rendering of legal advice, AI may support but does not replace the attorney's independent work.

5. Prohibited uses

The following uses are prohibited regardless of which tool is used:

1. Submitting any information that relates to a client representation into a tool that is not on the approved list.
2. Inputting any client information into any consumer or free-tier AI tool, including personal accounts on ChatGPT, Claude, Gemini, or similar services. Exception: if the provider's enterprise terms have been reviewed and approved under Section 3, that tool is no longer consumer-grade for firm use.
3. Relying solely on AI output to render legal advice, negotiate a client matter, or execute any task requiring the exercise of professional judgment.
4. Submitting AI-generated content to any court, arbitrator, or tribunal. All such filings must complete the verification protocol in Section 8 first.
5. Using AI to make or recommend an adverse decision about a person, including hiring, firing, client intake, or fee allocation. All such decisions require independent human review of the AI output for bias, accuracy, and compliance with RPC 8.4(g).
6. Any use that would violate this policy, applicable rules of professional conduct, a client's engagement terms, or outside counsel guidelines.

6. Client data and confidentiality

Before inputting any client information into any AI tool, the attorney responsible for the matter must confirm three things:

1. The tool is on the approved list.
2. The approved use case covers the intended use.
3. The approved-tool review establishes that the tool does not retain or train on submitted content in a way that raises a material confidentiality risk. Alternatively, the client has given tool-specific informed consent under Section 7.

ABA Formal Opinion 512 provides that a client's informed consent is required before inputting client information into a self-learning AI tool whose output could lead to disclosure of that information. Opinion 512 also states that boilerplate engagement-letter language is not sufficient.

Where informed consent is required, the responsible attorney documents the following in the matter file:

1. Why the tool is being used.
2. Which categories of client information will be input.
3. Disclosure risks, including how disclosed information could be used against the client's interests.
4. Benefits of the proposed use.

Consent is recorded with the date, signatory, and the specific tool and use case authorized. The Firm's standard form is the Informed Consent Form.

Firm information is classified into four tiers for AI-use purposes. Before inputting any information into an AI tool, the responsible attorney classifies it under the highest applicable tier and selects a tool approved for that tier:

7. Client disclosure and communication

The Firm discloses its AI practices to clients as follows:

• Default engagement letter language. The Firm's standard engagement letter includes a plain-language paragraph describing the Firm's use of AI tools and inviting the client to request additional detail or to restrict AI use. The Firm also publishes a standalone Notice of AI Practices on its public website.
• On request. Any client question about whether or how AI was used on the client's matter is answered accurately and promptly.
• Tool-specific consent. Where Section 6 requires informed consent for a specific tool and use case, the consent language is provided to the client before use. Consent is documented as described in Section 6.
• Outside counsel guidelines. Where a client's engagement terms or outside counsel guidelines require disclosure of AI use, the Firm complies with those terms regardless of whether this policy would otherwise require it.
• Significant-decision consultation. Where AI output will influence a significant decision in the representation, such as litigation outcome evaluation, jury analysis, or key drafting judgments, the responsible attorney consults the client before relying on the output.

8. Verification of AI-assisted filings

Any document submitted to a court, arbitrator, or other tribunal that was drafted, researched, or substantively supported by an AI tool is subject to the following verification protocol before filing. The protocol applies whether the tool was general-purpose or legal-specific, and whether citations were generated by the tool or only summarized by it.

• Citations. Every legal citation, including case names, reporters, docket numbers, statutes, rules, and regulatory references, is independently verified against a primary source.
• Quotations. Every quotation attributed to a cited authority is independently verified against the cited authority.
• Holdings and procedural posture. Each cited case is independently confirmed against the primary source for holding, reasoning, and procedural posture, not against the AI tool's summary.
• Jurisdiction. Cited authorities come from the controlling jurisdiction. AI tools regularly blend jurisdictions in summaries without flagging the substitution.
• Doctrinal reasoning. Confirm that the legal test, elements, or standard applied is correctly stated and that the conclusion follows from it. AI-generated reasoning may sound coherent while applying a superseded or wrong-jurisdiction doctrinal frame.
• Bias and characterization. Cited authorities are reviewed for fair characterization: no selective omission of unfavorable holdings, no overstatement, no omission of controlling adverse authority.
• Formatting and procedural rules. Captions, signature blocks, page and word limits, font and spacing requirements, and any judge-specific procedural rules are confirmed against the operative local and chambers rules.
• Attorney sign-off. The verification is signed and dated by the responsible attorney, who identifies the AI tool and version used and the verification steps completed. The signed log is retained in the matter file.

Cost-aware caveat: where verification of AI-assisted output requires more time than original drafting would have required, AI is presumptively not the right tool for that task. The responsible attorney should reconsider before using AI-assisted drafting in that situation.

The Firm's standard form for the verification log is the Verification Log Template.

9. Data security and privacy

Use of AI tools must comply with the Firm's general data security and privacy obligations. Specifically:

• Devices. AI tools approved for confidential or higher-tier data may be used only on Firm-managed devices, or on personal devices that meet the Firm's device-management requirements. AI use for any tier above Internal on devices outside Firm management is prohibited.
• Access. Access to enterprise AI accounts is provisioned per individual; account credentials are not shared between users. Account access is removed promptly on departure or role change.
• Network. AI tools are accessed only over networks that meet the Firm's general security expectations (encrypted connection, no public-WiFi use without VPN for Confidential or Highly Sensitive tier work).
• Logging and retention. Where an enterprise AI tool offers chat-history controls, the Firm configures retention to the minimum required for legitimate operational use. Personal accounts that retain chat history indefinitely are not approved tools regardless of how they are accessed.
• Accidental disclosure. Any inadvertent input of client information, privileged material, or other sensitive content into an AI tool is treated as a potential incident under Section 13. Staff must promptly notify [Managing Partner / AI Committee]. Do not wait to assess severity before reporting.
• Vendor security posture. Tools approved for Confidential or Highly Sensitive tier use must have recognized security evidence: SOC 2 Type II, ISO 27001, or equivalent. Document the evidence in the vendor file under Section 3.

10. Supervision and training

Under Rules 5.1 and 5.3, [Managing Partner / AI Committee / designated partner] owns firm-wide AI governance. Supervising attorneys remain responsible for the AI-assisted work product of attorneys and staff they supervise. No AI-assisted work product leaves the Firm without review by a responsible supervising attorney.

The Firm operates on a human-in-the-loop principle. AI may assist with drafting, summarization, brainstorming, or research support. A qualified attorney must review all output, apply professional judgment, and make every final decision that affects a client matter. Responsibility for the final work product runs to the attorney, not the tool.

Attorneys and staff who use an approved AI tool on a client matter must notify the responsible supervising attorney before submitting AI-assisted work product for review. The notification identifies the tool used, the use case, and the verification status under Section 8. This ensures supervisory review is informed under Rules 5.1 and 5.3, not retrospective.

For third-party AI providers, the vendor review covers: reference checks and credentials; security policies and protocols; confidentiality terms; conflicts screening where applicable; and whether the provider retains or claims proprietary rights to submitted content. Document the review in the vendor file for each approved tool, using the Vendor Due Diligence Checklist.

11. Billing treatment

The Firm bills AI-assisted work in accordance with Rule 1.5 and ABA Formal Opinion 512. Specifically:

• Hourly matters. The Firm bills only for time actually expended, including time spent prompting the AI tool and time spent reviewing its output. Time saved by AI-driven efficiency is not billed.
• Flat and contingent matters. Where AI materially compresses the work contemplated by a flat or contingent fee, the responsible partner considers whether the fee remains reasonable and, if not, raises the question with the client.
• Tool costs as overhead. AI tools that function like standard office infrastructure (for example, AI features embedded in word processing or email) are treated as overhead and are not billed to clients.
• Tool costs as pass-through expenses. Per-matter or per-use AI tools are billed as expenses at actual cost. No surcharge applies unless separately agreed in writing with the client.
• Learning time. Time spent by Firm personnel learning to use an AI tool that the Firm will use regularly for clients is not billed to any client.

12. Competence and training requirements

Every attorney and staff member using an approved AI tool must complete training before use. Training must cover at minimum:

1. The tool's capabilities and known limitations.
2. The confidentiality posture of the tool, including the provider's data retention and training practices.
3. Prohibited uses.
4. The verification protocol in Section 8.

Completion is logged in each user's training file with the date and tool identified.

Training is refreshed whenever an approved tool is materially updated or whenever a new approved tool is added. Attorneys are expected to maintain ongoing technological competence consistent with Rule 1.1 and Comment [8], including through CLE or equivalent continuing education.

13. Incident response

An AI incident includes, without limitation:

1. Inadvertent input of client information into an unapproved tool.
2. Disclosure of client information through an AI tool's output.
3. Submission of an AI-generated misstatement to a court, opposing counsel, or third party.
4. Any provider-side breach, outage, or service disruption affecting firm data.

On identifying an incident, the person who identifies it must notify [Managing Partner / AI Committee] within 24 hours. The Firm then assesses four questions:

1. Whether client notification is required under Rule 1.4 or applicable state rules.
2. Whether bar reporting is required.
3. Whether remedial disclosure to a tribunal is required under Rule 3.3.
4. Whether any insurance notice obligation is triggered.

The incident, the assessment, and the response are documented.

14. Exceptions and waivers

Using a tool not on the approved-tools list, or using an approved tool outside its approved use case or data tier, requires advance written approval from [Managing Partner / AI Committee / designated partner]. An exception request must describe:

1. Proposed tool and version.
2. Use case and matter, if any.
3. Data classification of the information to be processed.
4. Duration of the exception.
5. Any compensating controls or matter-specific safeguards.

Approved exceptions are logged in the AI exception register with date, scope, approver, and expiration. At each annual review under Section 15, the exception register is scanned for patterns that may justify amending the standing approved-tools list.

15. Policy review and update cadence

This policy is reviewed and re-approved at least annually. Review is also triggered by any material change to: (a) the approved-tools list; (b) applicable rules of professional conduct or authoritative bar guidance; or (c) the Firm's practice areas. The current version date and approver are recorded at the top of the policy.

Policy adopted: [date]. Approved by: [name, title]. Version: [1.0].

How to use this template

1. Adapt the approved-tools list and assign tiers. Section 3 is the most firm-specific part. Inventory all AI tools currently in use, including shadow AI on personal devices. Classify each into Tier 1, Tier 2, or Tier 3 before drafting the approved list. Most firms find that tools they assumed were Tier 1 are actually Tier 2 once contract terms are reviewed.
2. Have the policy reviewed by your malpractice carrier or broker. Carriers are increasingly asking about AI at renewal. A policy reviewed before renewal is documentation on the record. A policy produced in response to a claim is not.
3. Get managing partner sign-off. Rules 5.1 and 5.3 assign responsibility to managerial lawyers. The policy should be adopted by the managing partner or the firm's governing committee, not delegated.
4. Train the firm before going live. Section 12's training requirements take effect when the policy does. Schedule the initial training session and log completion before the effective date.
5. Collect signed acknowledgments. Every attorney and staff member with access to an approved tool signs an Employee Acknowledgment Form before access is provisioned. The roster of signed acknowledgments is one of the artifacts a firm should be ready to produce at malpractice renewal.

State-specific annexes

ABA Formal Opinion 512 is the national baseline, not the ceiling. Several states have issued their own guidance. Firms should check their state's rules before finalizing a policy. Sources to read alongside Opinion 512:

• Florida Bar Ethics Opinion 24-1 (January 2024)
• North Carolina State Bar 2024 FEO 1 (adopted 2024)
• California State Bar Practical Guidance on Generative AI (November 2023)
• Pennsylvania and Philadelphia Joint Formal Opinion 2024-200 (2024)
• Texas, New York, Illinois, and other state opinions and practice guides that have issued since 2024

The state tracker publishes a current state-by-state view with primary-source citations for each opinion and court order. Where state guidance runs stricter than Opinion 512, the state rule controls.

Carrier-readiness note

The documentation this template produces is what malpractice carriers have started requesting at renewal: the written policy, the approved-tools list with vendor review, training records, a verification protocol, and an incident response procedure. Many carriers now include AI-specific questions on renewal applications. Some offer credits for documented AI governance. See carrier renewal documentation for the underlying carrier activity.

Next steps

The template above is the free version. The expanded, bundled version is the Carrier-Renewal Packet: state-specific annexes, vendor diligence for common legal AI tools, engagement-letter and informed-consent language, and a pre-filing verification form, all pre-customized for the renewal binder.

Last verified against ABA Formal Opinion 512: 2026-04-29.