6 min readSanitized Ai Team

AI Acceptable Use Policy for Accountants

AI GovernanceComplianceData PrivacyShadow AI

An AI acceptable use policy should make everyday decisions easier for accountants. It should tell employees which tools they may use, what information they may enter, when human review is required, and what to do when a new AI feature appears inside software the firm already uses.

A policy that only says “use AI responsibly” leaves too much room for interpretation. The most useful policies translate privacy, confidentiality, professional judgment, and security expectations into rules employees can apply before they paste a prompt or upload a file.

1. Define what the policy covers

Start with the people, tools, and activities covered by the policy.

It should apply to employees, partners, contractors, and other users handling firm or client information. It should also cover more than standalone chatbots. AI features can appear in accounting platforms, document systems, meeting tools, browser extensions, and productivity software.

2. Maintain a list of approved AI tools

Employees should know which AI environments have been reviewed by the firm.

For each approved tool, document the permitted purpose and data categories. A product may be appropriate for drafting with public information but not for payroll data, client financial statements, or confidential tax documents.

NIST’s AI Risk Management Framework is designed to help organizations manage AI risk according to context, while its Generative AI Profile extends that approach to generative AI risks.

Suggested policy language:

Use only firm-approved AI tools for firm or client work. Approval applies only to the purposes, features, data types, and integrations specifically authorized by the firm.

3. Set rules for personal and client information

The policy should identify information that requires special care.

For accountants, this can include bank details, tax identifiers, payroll records, employee information, customer data, financial statements, forecasts, contracts, and confidential business information.

Canadian privacy regulators recommend using anonymized, synthetic, or de-identified information instead of personal information when the real data is unnecessary. They also state that sensitive or confidential personal information should be entered into generative AI prompts only where authorized.

Suggested policy language:

Do not enter personal, confidential, or client information into an AI system unless the system and specific use have been approved for that information. Remove or replace unnecessary information whenever possible.

4. Prohibit a few high-risk behaviours explicitly

Some rules should not depend on interpretation.

Prohibit entering passwords, authentication tokens, private keys, recovery codes, or other credentials into AI prompts. The policy can also prohibit bypassing security controls, using personal AI accounts for confidential client work, or connecting AI tools to firm repositories without approval.

5. Require meaningful review of AI output

AI can assist with drafting and analysis, but the accountant remains responsible for the work.

IESBA’s current technology guidance says professional accountants must assess whether technology outputs are fit for purpose, understand relevant limitations and assumptions, consider data quality and potential bias, and determine the appropriate extent of reliance. It also warns about automation bias and states that professional responsibility cannot be delegated to a machine.

A rewritten internal email may need a basic accuracy check. Technical accounting analysis, calculations, client communications, or information used in a professional decision should receive stronger independent verification.

Suggested policy language:

AI-generated output must be reviewed before it is relied upon. Material facts, calculations, technical claims, and professional conclusions must be independently verified using appropriate source information.

6. Address file uploads and integrations separately

Uploading an entire file creates more exposure than typing a short prompt.

Before uploading spreadsheets, PDFs, or other documents, employees should remove irrelevant information and check for hidden sheets, comments, metadata, or unrelated pages where appropriate.

Connected applications need separate approval. An AI assistant linked to email, cloud storage, or an accounting platform may access much more information than employees manually provide. Use least privilege and approve only the data sources required for the intended workflow.

7. Explain retention and recordkeeping

Employees should not assume that information disappears when a chat window is closed.

The firm should determine what prompts, files, outputs, and logs need to be retained, and what should be deleted when no longer required. Canadian privacy regulators recommend appropriate retention schedules for personal information in prompts and outputs and limiting secondary uses.

Higher-risk uses may also require records showing the purpose, tool, important inputs, review performed, and final decision.

8. Create a simple incident rule

Employees need a clear response when something goes wrong.

Examples include accidentally uploading client information to an unapproved service, discovering that an AI integration has broader access than expected, or finding that an AI-generated result used in work was materially incorrect.

Suggested policy language:

Promptly report suspected AI-related privacy, confidentiality, security, or accuracy incidents through the firm’s designated reporting process.

9. Require training and periodic updates

IESBA’s July 2026 guidance on emerging technologies emphasizes appropriate technological literacy and ongoing learning as technology develops. Training should cover approved tools, prohibited data, human review, incident reporting, and new features.

The policy should be reassessed when the firm adopts a new AI tool, enables a significant integration, changes the data being processed, or sees a material change in vendor capabilities.

Keep the policy usable

An accountant should be able to answer six questions before using AI:

  1. Is this tool and account approved?
  2. Is this use case allowed?
  3. Does the prompt or file contain personal or confidential information?
  4. Can I remove information the AI does not need?
  5. How must I verify the output?
  6. Do I need additional approval or documentation?

A strong AI acceptable use policy for accountants does not try to predict every future AI feature. It establishes clear boundaries that continue to work as tools change.

The policy should make safe AI use easier, not simply create another document employees sign once and forget.

Every rule in an acceptable use policy shares a timing problem: once a prompt is submitted, it cannot be recalled. A client tax identifier pasted into a drafting request, or a payroll spreadsheet uploaded to an unapproved assistant, has already been shared the moment the employee presses enter. This is the principle Sanitized Ai is built on, catching and redacting personal and client information before the prompt reaches the tool, so the policy's rules on prohibited data are enforced at the moment they matter.

This quarter, take the six questions above and test them against the one AI workflow your team uses most, whether that is summarizing client documents or drafting emails. Note which identifiers the AI never needed, write their removal into the policy as the default rather than the exception, and walk your staff through the result. If you want to see how redaction can happen before submission rather than after, request a demo and review your own policy with us.

See how Sanitized Ai stops sensitive data from leaving the prompt box. Writing your own rules instead? Start from our free AI acceptable use policy generator.