All templates

Law & Legal AI Policy Generator and Template

Create a starting policy for law firms and in-house legal teams. Set matter, tool, and data boundaries; preserve lawyer supervision; and define how research, drafting, client communication, billing, and incidents are reviewed before adoption.

Updated:

A legal AI policy starts with professional duties

Legal work combines confidential client information, duties to current and former clients, court and tribunal rules, deadlines, professional judgment, and records that may be discoverable. A product being marketed as an AI assistant does not decide whether a particular matter, input, output, or use is permitted.

Use this template as a decision aid for a law firm or in-house legal function. Identify the responsible lawyer or legal-operations owner, the jurisdictions and tribunals in scope, the matter restrictions that apply, and the approval route for a new tool or use. A legal policy cannot replace the professional judgment or local advice required for a particular engagement.

  • State whether the rule covers lawyers, paralegals, legal operations, support staff, contractors, outside counsel, and vendors.
  • Separate firm-wide or department-wide defaults from matter-specific instructions, engagement terms, client directions, and court or tribunal rules.
  • Name who can approve a tool, a data category, a matter, an exception, and an external communication.
  • Record a version, effective date, review owner, escalation contact, and the jurisdictions whose rules must be checked.

Client confidentiality, privilege, and work product

Do not put client information, matter strategy, personally identifiable information, trade secrets, health or financial information, credentials, or nonpublic work product into an AI service unless the specific tool, account, data, purpose, retention setting, access model, and matter use have been approved. Minimize the prompt and use the firm or client-controlled environment where one is required.

Confidentiality, attorney-client privilege, and work-product protection are legal and jurisdiction-specific questions. A label such as “private,” an enterprise plan, encryption, or a vendor promise does not guarantee that a communication or generated material is protected, that a waiver cannot occur, or that an output is work product. Confirm the facts and applicable law with qualified counsel and follow the client’s instructions.

Do not use a shared conversation, consumer account, unapproved plug-in, or connected workspace for a matter merely because it is convenient. Treat prompts, uploaded files, conversation history, generated drafts, logs, feedback, and metadata as potentially sensitive records. Preserve only what the matter and records rules require, and document an approved deletion or retention instruction where appropriate.

  • Default to no client or matter data in a tool until the tool and the intended use are separately approved.
  • Check training-on-input settings, retention, human access, subprocessors, cross-border transfers, account separation, and incident notice.
  • Keep matter boundaries intact: do not combine clients, adverse parties, restricted teams, or unrelated workspaces in a prompt or retrieval index.
  • Escalate a possible disclosure, conflict, privilege concern, or loss of control immediately through the firm’s incident and supervising-lawyer process.

Approved tools, vendors, and matter-level permissions

Approve tools for a defined use rather than approving a brand in the abstract. Vendor review should cover the model and features actually enabled, terms and permitted uses, data ownership, retention and deletion, security, access controls, subprocessors, locations, confidentiality commitments, audit information, service changes, and incident response. Re-review material changes instead of assuming an earlier approval remains current.

A tool approved for public legal research may not be approved for a draft pleading, a client email, a privileged memo, or a matter with protective-order restrictions. Matter partners, general counsel, client instructions, outside-counsel guidelines, and court orders can impose stricter conditions. The policy should make the narrowest controlling rule easy to find.

Use an approval record that identifies the tool, account or workspace, legal purpose, data classes, permitted matters, prohibited uses, accountable owner, reviewer, expiry or review trigger, and fallback when the service is unavailable. Never place passwords, API keys, access tokens, or confidential credentials in a prompt.

Research, drafting, and lawyer supervision

A lawyer remains responsible for work product and advice that relies on AI assistance. Verify legal research against authoritative primary sources, including the current statute, regulation, rule, order, opinion, or official guidance. Read the cited authority and confirm jurisdiction, procedural posture, effective date, amendments, subsequent history, quotations, pinpoint citations, and the proposition for which it is offered. Do not treat a fluent answer, citation, headnote, or database summary as authority without checking it.

Review generated contracts, pleadings, briefs, letters, discovery work, client advice, translations, timelines, coding, and calculations for accuracy, omissions, invented authorities, confidentiality, conflicts, privilege, bias, tone, accessibility, and fitness for the matter. The supervising lawyer must have enough understanding of the tool and enough time and authority to make a meaningful review; “human in the loop” is not a substitute for supervision.

Define when AI assistance must be recorded in the matter file and how a reviewer can reproduce or understand material assistance without unnecessarily preserving sensitive prompts. Follow litigation holds, professional-responsibility duties, client directions, records schedules, and local court requirements. Do not allow an automated score to decide credibility, sanctions, eligibility, settlement authority, or a client’s legal position.

  • Research check: identify and read the primary source, then record the jurisdiction, date checked, and any controlling updates.
  • Draft check: compare every material assertion, quotation, citation, defined term, number, deadline, and cross-reference against the matter record.
  • Supervision check: assign a named lawyer with subject-matter authority, an escalation route, and sufficient time to review before reliance or filing.
  • Filing check: confirm court, tribunal, judge, client, and opposing-party requirements before using or submitting AI-assisted work.

Write your policy

Customize this template

Client communications, consent, fees, and reporting

Decide when the engagement, client instruction, applicable professional rule, court order, or material change in the service requires client notice, discussion, or consent. Explain the intended use in plain language, the information involved, material risks, safeguards, alternatives, and who remains responsible. Do not imply that a client has consented merely because an engagement letter is silent or because a tool is common in the market.

Set a billing rule before use. Record actual time and work performed accurately, distinguish human legal judgment from automated assistance, and do not charge a client for invented work, duplicate effort, or an unapproved markup. Follow the engagement, client billing guidelines, fee rules, and any requirement to explain or obtain agreement about technology charges. The policy should identify who reviews invoices and narratives for accuracy.

Report suspected disclosure, hallucinated authority, inaccurate advice, missed deadline, unauthorized matter access, vendor incident, conflict, court-rule concern, or client complaint promptly. Preserve necessary evidence securely, stop the affected use when appropriate, notify the supervising lawyer or incident owner, and follow applicable notice and remediation procedures. Record what was checked and what decision was made without circulating sensitive material unnecessarily.

  • Client communication is context-specific: assess confidentiality, materiality, informed consent, engagement terms, client policy, and the governing jurisdiction.
  • Before filing or sending, a lawyer—not the tool—owns the final authority, factual accuracy, citations, tone, and disclosure decision.
  • Use a billing and time-recording instruction that reflects actual work and the applicable client or professional requirements.
  • Give every user one clear route for questions, urgent incidents, conflicts, and possible court or tribunal issues.

Jurisdiction, court rules, and adoption review

This template is intentionally not worldwide legal guidance. ABA Formal Opinion 512 and its summary are U.S. ethics guidance grounded in the ABA Model Rules; they are not a substitute for the rules adopted in a particular U.S. jurisdiction, a client’s requirements, or advice about a specific matter. The Solicitors Regulation Authority source is relevant to solicitors and firms it regulates in England and Wales. It is not a general statement of law for every country, regulator, lawyer, or legal profession.

Before adoption, have qualified local reviewers reconcile this draft with professional-conduct rules, confidentiality and data-protection duties, legal-aid or public-sector requirements, engagement letters, client outside-counsel guidelines, discovery and records obligations, and the rules of each court or tribunal in scope. Check current versions and local interpretations; AI guidance and vendor features can change.

A generated document does not establish compliance, preserve privilege, or receive legal review. Adopt only after the responsible lawyers resolve open decisions, test realistic matters, train users, publish approved-tool and reporting information, and schedule review after material changes, incidents, new authorities, or new client requirements.

  • US practice: identify the state or other jurisdiction’s adopted rules and applicable court or tribunal requirements; ABA materials are reference guidance, not a universal rulebook.
  • England and Wales: determine whether SRA standards and guidance apply to the people and firm in scope; check other relevant requirements separately.
  • Other locations: obtain qualified local review rather than extrapolating from either ABA or SRA material.
  • Approval is conditional on implementation: an adopted policy is not evidence that its controls, training, or vendor settings are operating.

Example wording

Matter and research approval record

Approved use: compare the structure of public authorities in the managed research workspace for the Acme employment matter. Data: public sources only; no client communications, strategy, personal data, credentials, or restricted filings. Conditions: the responsible lawyer reads and verifies each current primary authority, checks jurisdiction and subsequent history, records the date checked, and approves every citation before it appears in advice or a filing. Not approved: client documents, privileged material, automated legal conclusions, or filing without lawyer review.

Client communication and billing clause

Before using AI in a material client-facing workflow, the matter lawyer assesses the engagement, client instructions, confidentiality and consent considerations, court requirements, and the tool’s approved data settings. Any notice or consent describes the intended use and safeguards accurately. Time records and invoices reflect actual human work and applicable fee rules; no charge is made for fictitious or duplicate work. Suspected disclosure, inaccurate authority, or unauthorized use is reported promptly to the supervising lawyer and incident owner.

Sample policy

Example Legal Team Legal AI Use Policy

The names are placeholders. Review the rules and fill in the missing details before using this policy.

Read the sample policy
Audience
Law & Legal
Scope
organization
Owner
To be designated
Status
Draft
Version
0.1
Effective date
To be confirmed
Review date
To be scheduled

Summary

  • Law & Legal policy. Status: Draft. Version: 0.1.
  • Activity permissions: Allowed: 0; Allowed with conditions: 0; Approval required: 0; Prohibited: 8.
  • This draft approves no tools. Designate approved tools before use.
  • Secrets are never permitted in general-purpose AI tools; sensitive data requires exact data-type, tool, and use approval.
  • Lawyer responsibility, supervision, primary-source verification, confidentiality, privilege, work product, matter access, court-rule checks, and actual-time billing remain mandatory.

1. Document control and scope

This policy applies across Example Legal Team to people and work covered by its authority. Owner: designated policy owner. Status: Draft. Version: 0.1. Region, if supplied, is review context only and does not establish local legal compliance.

2. Purpose and approach

This policy sets the rules for AI use and assigns accountability. Use is restricted. An activity is permitted only as stated in the activity matrix and subject to every listed condition.

3. Definitions

“AI tool” means software that generates, transforms, ranks, or analyzes content using an AI model. Legal assistance may organize or transform material while preserving a lawyer's independent professional judgment. Substantive generation creates or materially rewrites legal analysis, factual assertions, advice, drafts, or communications; generated text is not legal authority. “Approved tool” means a tool designated for a particular use and data type through the applicable approval process; familiarity or public availability is not approval.

4. Roles and responsibilities

The designated policy owner maintains this policy and coordinates its review. the designated approver decides requests to approve tools, exceptions, and uses within their authority. the person responsible for the work checks outputs before anyone relies on or releases them. Lawyers retain independent professional judgment and responsibility for legal work. A supervising lawyer assigns and reviews work appropriately, verifies authorities and facts, and ensures that use complies with professional duties, client instructions, matter terms, and applicable jurisdictional rules.

5. Approved tools and approval process

No tools are currently designated as approved. Before use, the designated approver must approve the vendor, tool, account or deployment, intended activity, data types, matter access, retention, and applicable terms. Public availability or prior use does not constitute approval.

6. Permitted-use matrix

The activity matrix is part of this policy. For each activity, its permission, required legal safeguards, additional conditions, and clarification apply. Required legal safeguards are controlling minimum requirements. User-added conditions and clarifications may add requirements but cannot waive, narrow, contradict, or override a required legal safeguard; where wording conflicts, the required safeguard controls. All applicable controls must be met.

Activity permissions and required conditions
ActivityPermissionConditions and clarification
Legal researchProhibitedNo activity-specific conditions. The policy's data and human-review safeguards still apply.
Drafting contracts and agreementsProhibitedNo activity-specific conditions. The policy's data and human-review safeguards still apply.
Drafting pleadings, motions, and briefsProhibitedNo activity-specific conditions. The policy's data and human-review safeguards still apply.
Matter document reviewProhibitedNo activity-specific conditions. The policy's data and human-review safeguards still apply.
Client communicationProhibitedNo activity-specific conditions. The policy's data and human-review safeguards still apply.
Court submissions and filing preparationProhibitedNo activity-specific conditions. The policy's data and human-review safeguards still apply.
Billing and administrative workProhibitedNo activity-specific conditions. The policy's data and human-review safeguards still apply.
Internal knowledge management and trainingProhibitedNo activity-specific conditions. The policy's data and human-review safeguards still apply.

7. Prohibited uses

The matrix prohibits: Legal research, Drafting contracts and agreements, Drafting pleadings, motions, and briefs, Matter document review, Client communication, Court submissions and filing preparation, Billing and administrative work, and Internal knowledge management and training. Do not provide autonomous legal advice, make an independent legal judgment on a client's behalf, impersonate a lawyer, make an unauthorized filing or court submission, fabricate authority, citations, quotations, facts, time entries, or billing records, or use AI in a way that creates an unauthorized-practice, conflicts, confidentiality, privilege, work-product, or supervision concern. Do not use AI where law, contract, professional duty, or a controlling organizational rule prohibits it.

8. Data, privacy, confidentiality, and intellectual property

The only data categories selected for possible use are: public. Selecting a category does not by itself permit its use. Matter data may be entered only when the specific tool, vendor terms, account or deployment, matter access, and proposed use are approved for that exact data type. Protect client confidentiality, attorney-client privilege, and attorney work product; segregate matters and limit access to the people and vendors authorized for that matter. No tool or workflow guarantees privilege, confidentiality, or a particular legal result. Credentials, passwords, access tokens, private keys, authentication codes, and other secrets must never be entered into general-purpose AI tools. Inputs and outputs remain subject to applicable professional duties, client and matter terms, access, retention, discovery, records, and intellectual-property requirements. If unsure, stop and ask the responsible lawyer or policy owner.

9. Output verification and human accountability

A person remains accountable for AI-assisted work. Before using or releasing it, check factual accuracy, calculations, citations, originality, bias, security, rights, and suitability for the intended audience. AI output is neither an authority nor a verified source. The responsible lawyer must independently exercise legal judgment and verify every authority, citation, quotation, factual assertion, deadline, calculation, and current legal status against primary sources. A supervising lawyer must review work at an appropriate level before reliance, client communication, filing, or submission. AI must not give autonomous legal advice, decide a matter, or file with a court.

10. Disclosure and attribution

Disclose AI assistance when it creates, materially rewrites, analyzes, or otherwise makes a substantive contribution. A declaration must accurately describe the use. The following example is not a declaration on the user's behalf. Client communication, consent, or disclosure must be handled only where applicable under professional rules, client instructions, or matter terms; it must not imply that a lawyer delegated professional judgment to an AI system.

11. Legal practice and matter rules

Lawyers must follow the professional rules and court or tribunal requirements applicable to each matter; guidance varies by jurisdiction. Research, drafting, review, communication, filing, and administrative use must remain within the responsible lawyer's authority and supervision. Verify authorities, citations, quotations, facts, deadlines, calculations, and current status against primary sources. Confirm court filing requirements and local rules before any submission. Do not give autonomous legal advice or make an autonomous filing. Protect client confidentiality, attorney-client privilege, work product, conflicts information, and matter segregation. Check approved-tool, vendor-term, account, and matter-access requirements before use. When billing by time, record only actual time and work performed, never fictional hours or claimed time saved; fixed or value-based fees and AI costs must follow agreed terms, reasonable-fee requirements, and applicable rules. Review current jurisdiction-specific guidance, including ABA Formal Opinion 512 (July 29, 2024) and the SRA Misuse of AI notice (Aug 17, 2026), where relevant; these references do not replace the rules applicable to a matter.

12. Review, reporting, and exceptions

Report concerns to the designated policy owner through the usual organizational channel. Report suspected data exposure, unsafe output, rights concerns, or material misuse promptly. The responsible lawyer must preserve relevant prompts, outputs, sources, approvals, and matter records as appropriate; assess the concern under applicable professional duties and matter terms; and take corrective action through the authorized supervisory, client, court, or firm process. Do not treat an AI output as authority or evidence without human verification. Exceptions must be written, limited in scope and duration, approved by the designated approver, and retained by the policy owner.

13. Training, communication, and acknowledgment

Before covered users begin AI use, the policy owner will provide onboarding on the matrix, approved tools, data limits, disclosure, verification, and reporting. Training must cover lawyer responsibility, supervision, jurisdiction-specific guidance, authority and citation verification, confidentiality, privilege, work product, matter access, court and local-rule checks, client communication, and accurate billing. Covered users will be asked to acknowledge that they received and understood the policy; acknowledgment does not waive rights or replace training.

14. Review and version history

The policy owner will review this policy annually and after a material incident, tool change, or change in applicable obligations. Changes require the designated approval process and notice to affected people. Confirm the effective date before adoption. Schedule the next review date. Version history: Version 0.1 | Status: Draft | Change: Initial draft.

Decisions to confirm

  • Name the policy owner.
  • Name the role or person authorized to approve tools and exceptions.
  • Designate tools, deployments, intended uses, and approved data types.
  • Set an effective date before adoption.
  • Schedule the next review date.

Implementation checklist

  • ☐ Confirm and name the policy owner.
  • ☐ Confirm jurisdiction-specific professional guidance, lawyer supervision, matter access, confidentiality, privilege, work product, primary-source checks, filing rules, client communication, and actual-time billing controls.
  • ☐ Review the sensitive-data rules for each proposed data type and use.
  • ☐ Designate approved tools before permitting AI use.
  • ☐ Resolve 5 open decisions before adoption.
  • ☐ Complete the authorized approval process; this document remains Draft.
  • ☐ Communicate the matrix, safeguards, disclosure rules, and reporting route.
  • ☐ Schedule the annually review.

Acknowledgment

I received the Example Legal Team Legal AI Use Policy and had an opportunity to ask questions. I understand where to find the current activity matrix and how to report concerns. This acknowledgment does not replace training or waive any rights. Name: ____________________ Signature: ____________________ Date: ____________________

This policy draft is not legal advice and does not make a legal or compliance determination. It does not guarantee confidentiality, attorney-client privilege, or attorney work-product protection. Required legal safeguards control over conflicting custom text. Local professional legal review is required before adoption; review the rules, client or matter terms, and court or tribunal requirements applicable to each matter.

Frequently asked questions

Does using a private or enterprise AI tool preserve privilege?

Not automatically. Privilege, confidentiality, and work-product treatment depend on facts, jurisdiction, purpose, people involved, tool terms, access, retention, and the applicable rules. Do not promise protection based on a product label; obtain qualified review and follow client and matter instructions.

Can a lawyer use AI for legal research or drafting?

A firm may permit defined uses after tool, data, matter, and supervision review. A lawyer remains responsible for checking primary authority and every material output, including citations, quotations, deadlines, and factual assertions. Court, client, and local professional rules may impose additional conditions.

Do clients always need to be told about AI use?

There is no single worldwide answer. Assess the engagement, client instructions, materiality, confidentiality and informed-consent duties, governing jurisdiction, court rules, and the service’s risks. The policy should identify when notice, discussion, or consent is required and record the decision accurately.

How should AI-assisted time and fees be reported?

Record actual time and work performed, follow the engagement and client billing rules, and do not charge for fictitious, duplicate, or unapproved work or markup. State who reviews narratives and how technology costs are handled where that matters. Local fee rules and client guidelines control.

Are the ABA and SRA sources worldwide legal requirements?

No. ABA Formal Opinion 512 is U.S. ethics guidance associated with the ABA Model Rules, not a universal law or a substitute for a jurisdiction’s adopted rules. The SRA source applies to solicitors and firms within the SRA’s remit in England and Wales. Other lawyers and organizations need qualified local review.

Sources and references

These drafts have not been reviewed by a lawyer or subject-matter expert. Have your legal or policy team review them before use. They are not legal advice or proof of compliance.

NIST AI Risk Management Framework

NIST describes its AI Risk Management Framework as voluntary guidance for managing AI-related risks and incorporating trustworthiness into AI use.

Use it as a review reference when assigning owners and considering risks, safeguards, and ongoing oversight. This generator does not perform a NIST assessment.

UNESCO: Guidance for generative AI in education and research

UNESCO’s guidance emphasizes a human-centred approach, data privacy, and age-appropriate use of generative AI in education and research.

For school, university, and classroom policies, review student data, age restrictions, access, and the learning purpose of each permitted activity with the appropriate institutional owners.

Pangram: Useful AI policies for educators

Pangram’s classroom-policy article compares blanket restrictions with a tier-based approach that allows different levels of AI assistance for different assignments.

Use the examples to discuss classroom boundaries, then choose your own permissions in the generator. These approaches are options, not a requirement to adopt a particular rule.

ABA Formal Opinion 512: Generative Artificial Intelligence Tools (July 29, 2024)

ABA Formal Opinion 512 (July 29, 2024) provides U.S. ethics guidance associated with the ABA Model Rules on lawyers’ use of generative AI tools.

Use it as a U.S. Model Rules reference, then check the rules adopted by the relevant jurisdiction, client instructions, and matter requirements.

ABA summary: ABA issues first ethics guidance on AI tools

The ABA summary describes the organization’s first formal ethics guidance on generative AI tools and its focus on existing lawyer duties.

This is U.S. professional-ethics context, not a worldwide legal requirement or a substitute for local qualified review.

SRA: Misuse of AI warning (published 17 August 2026)

The Solicitors Regulation Authority warning on misuse of AI was published 17 August 2026 for solicitors and firms within the SRA’s remit.

Apply it only where SRA standards and guidance govern; it is not a statement of law for every jurisdiction or legal profession.

References do not imply endorsement.

Create your AI policy