Governance starter
A useful policy says which tool, which account, which data, which workflow, and which reviewer. It also gives staff a fast path to propose something new.
Published by PrivateStride · Last updated August 11, 2026
Important context
This is a policy-design starter, not a finished policy or legal advice. Your firm should tailor it to its services, clients, contracts, professional duties, employment practices, information security program, records rules, and applicable law.
Before drafting
A policy cannot repair an unknown data flow or an unusable approved tool.
Start with the AI inventory, information classes, intended workflows, system diagrams, vendor review, risk assessment, and professional-review requirements. Involve the firm’s Qualified Individual, legal and privacy advisers, IT provider, engagement leaders, records owner, and people responsible for training.
Keep the employee-facing policy short. Put changing details such as approved product names, settings, owners, and workflows in a controlled AI register that can be updated without rewriting the entire policy.
Start by mapping likely shadow AI usePolicy structure
The sample language is intentionally general. Replace it with decisions your authorized reviewers have approved.
Policy section
State why the policy exists, who it covers, and which AI systems count. Include stand-alone chatbots, meeting tools, browser extensions, embedded software features, APIs, models running locally, and AI used by contractors when they handle firm work.
Starter language: This policy governs the use of AI systems for firm business by personnel and contractors. AI may support work only when the service, account, data, and workflow are approved under this policy.
Policy section
Maintain a controlled list with the exact product, account or workspace, permitted data classes, approved workflows, owner, settings, and approval date. A familiar vendor name is not enough.
Starter language: Users may use only the systems and account types listed in the firm’s AI register. Personal accounts and unlisted AI features may not be used for firm work.
Policy section
Translate the firm’s information classification into simple AI instructions. State what may enter each approved environment and what always requires specific review.
Starter language: Client, taxpayer, employee, credential, and other restricted information may enter an AI system only when the written register expressly permits that data class and workflow.
Policy section
Make clear that AI output is a draft. Assign review for facts, calculations, sources, tax positions, confidential information, bias, tone, and records treatment before reliance or delivery.
Starter language: The responsible professional must verify AI-assisted work to the same standard as other work. AI does not approve a conclusion, sign a deliverable, or replace professional judgment.
Policy section
Name uses the firm will not allow, such as entering restricted data into an unapproved system, exposing credentials, hiding AI use from a required reviewer, auto-sending unreviewed output, or delegating a decision that must remain with a professional.
Starter language: Users may not bypass access controls, connect an unapproved plug-in, disable required logging, or represent unverified AI output as established fact.
Policy section
Explain which prompts, inputs, outputs, approvals, corrections, and incidents become firm records. Give staff a nonpunitive path to report a mistake or unapproved use quickly.
Starter language: Report accidental disclosure, suspicious output, unexpected system behavior, or unapproved use immediately through the firm’s security process. Prompt reporting supports containment and review.
Policy section
Provide a short intake form and name the decision owner. Require the purpose, data, service, flow, professional reviewer, expected value, and proposed test. Time-box exceptions and document conditions.
Starter language: A new AI use requires approval before production data is used. Material changes to the provider, model, integration, purpose, or data path trigger review again.
Policy section
Train with real firm scenarios and verify understanding. Apply existing employment and professional policies consistently. Review the AI policy on a defined schedule and after significant changes or incidents.
Starter language: Users receive role-based instruction before access and periodic refreshers. The firm reviews this policy and AI register at least annually and when material changes occur.
AI register
One row per approved workflow gives users, administrators, reviewers, and auditors a common record.
| Register field | What to record | Why it matters |
|---|---|---|
| Service and account | Exact product, edition, workspace, owner, and administrators | Prevents a brand-level approval from spreading to other products |
| Workflow and purpose | Task, engagement context, users, input, output, and intended use | Gives reviewers the facts behind the approval |
| Permitted data | Allowed classes, explicit exclusions, masking, and test-data rules | Lets staff decide before they upload |
| Controls | Access, MFA, retention, logging, training, review, and records handling | Connects the policy to the deployed system |
| Approval | Business, legal, privacy, security, tax, and records decisions as applicable | Shows who accepted which conditions and limitations |
| Lifecycle | Approval date, review date, changes, incidents, suspension, and exit | Keeps a once-correct decision from silently going stale |
Launch checklist
Use examples from tax, audit, advisory, administration, and client service. Include a safe response when a user is uncertain: stop, preserve the information, and ask the named owner before proceeding.
See how PrivateStride builds adoption habitsSupporting frameworks
Governance, data mapping, risk assessment, technical controls, testing, incident response, and leadership reporting need to agree with the written rule.
Primary sources: NIST’s voluntary AI Risk Management Framework and Generative AI Profile offer governance and risk-management considerations. The FTC’s Safeguards Rule guide describes information-security program elements for covered financial institutions. The IRS maintains the Section 7216 Information Center. See our claims methodology for how we separate measured observations from projections.
Keep reading
These guides cover the decisions that sit next to this one.
Define the product, data, purpose, and review behind each policy decision.
Read the guideConnect the AI register to the written security program.
Read the guideBring tax return information questions into workflow approval.
Read the guideCapacity & AI risk assessment
In 30 minutes, we identify your highest-value workflows, likely shadow-AI exposure, and the controls a private AI program would need. The findings are yours to keep.
