The IRS did not tell tax firms to avoid AI. It told them to stay responsible.

The IRS Office of Professional Responsibility did not tell tax professionals to step away from AI. Its June 2026 Introductory Guidelines for Responsible AI Use in Federal Tax Practice acknowledges that AI is widespread in modern tax and accounting offices and can provide real value.

The message is more practical: using AI does not transfer the professional’s responsibility to the model.

The guidance does not create a new regulation. It applies familiar duties—accuracy, competence, confidentiality, firm procedures, fair fees, and sound written advice—to AI-assisted practice.

For a firm, the question is no longer just “Do we have an AI policy?” It is “Could we reconstruct what happened on this engagement and show where professional review occurred?”

The duties are familiar, even if the technology is new

The document organizes its analysis around existing authority:

• Section 10.22 — diligence as to accuracy: verify AI-generated facts, citations, and calculations.
• Section 10.27 — fees: do not bill for manual time not spent or double bill for AI-assisted tasks; account for efficiency honestly.
• Section 10.35 — competence: understand the technology’s operation, limitations, and risks well enough to evaluate its output.
• Section 10.36 — procedures to ensure compliance: establish adequate firm procedures, train staff, vet providers, and document the processes.
• Section 10.37 — written advice: verify facts, law, assumptions, projections, and formulas rather than relying blindly on AI output.
• IRC 6713 and 7216, plus Circular 230 Section 10.51(a)(15): protect taxpayer return information from unauthorized use or disclosure.

The conclusion is straightforward: AI may assist the professional. It does not take responsibility from the professional.

Build a record that can answer what happened

A firm should be able to reconstruct an AI-assisted procedure without relying solely on the original chat.

For each material use, record:

Field	What to capture
Firm user	Who initiated the request
Client and engagement	Which taxpayer or company and which service
AI system	Product, provider, and material configuration
Connection	How the system reached client data
Instruction	The request or procedure given to the AI
Data scope	Reports, records, documents, or tax information accessed
Output	Material conclusion, draft, exception list, or proposed action
Verification	Sources checked, calculations recomputed, and records reviewed
Reviewer	Qualified person who evaluated the output
Approval	Who approved advice, filing content, or accounting changes
Result	What was delivered, changed, posted, or rejected
Time	Relevant request, review, approval, and completion timestamps

Not every low-risk use needs the same detail. Drafting a meeting agenda is different from preparing an adjusting entry that affects a tax return. The firm should define record depth by risk and engagement.

A policy cannot tell you what happened on one engagement

An AI policy answers questions such as:

• Which tools may staff use?
• What client data may enter each tool?
• Which tasks require a qualified reviewer?
• Which providers has the firm approved?
• What training is required?
• How should an error or disclosure incident be handled?

An activity record answers different questions:

• Which tool did this employee use for this client?
• What data did it access?
• What did it propose?
• What did the reviewer verify?
• What was approved?
• What reached the books, advice, or return?

The guidance calls for both procedures and documentation. One does not replace the other.

The audit trail often begins before the tax software

QuickBooks is often upstream of the tax return.

In QuickBooks Online Accountant, Prep for taxes lets a practitioner review annual balance sheet and P&L totals, inspect transactions, make adjusting journal entries, map accounts to tax lines, and export the result to ProConnect or CSV.

AI-assisted procedures may affect that chain before data reaches the tax software:

• clearing Uncategorized Income or Ask My Accountant;
• reviewing year-end account balances;
• preparing depreciation or accrual entries;
• reclassifying transactions;
• reviewing 1099 vendors;
• investigating unusual expenses;
• assembling the adjusted trial balance.

The return may be filed from another system. The accounting records still supply many of the amounts and classifications used to prepare it.

QuickBooks records the change, but not the whole decision

QuickBooks Online’s audit log is valuable. Intuit says it records account activity, user sign-ins, settings changes, and edits to records. It also keeps transaction-level audit history showing who changed a transaction, when, and what changed.

For third-party applications, Intuit documents an important attribution detail: when a connected app sends data or changes an existing record, the event can appear under System Administration.

That entry proves a change occurred. It may not identify:

• the staff member who initiated the AI request;
• the instruction given to the assistant;
• the records the assistant read before proposing the change;
• the reviewer who approved it;
• related actions in other companies;
• rejected or failed proposals.

Keep the QuickBooks audit log. Add a workflow-level record for the context it was not designed to capture.

What this looks like during year-end cleanup

Suppose a firm asks an AI assistant:

Review the year-end trial balance and general ledger. Identify unusual balances, prepare likely reclassifications, and list the items that require client support. Do not post changes.

A defensible record could contain:

1. Scope: Client company, fiscal year, accounts reviewed, and authorized data sources.
2. Instruction: The exact procedure and the no-posting boundary.
3. Findings: Unusual balances and the transactions supporting each finding.
4. Proposals: Draft reclassifications, separated from confirmed facts.
5. Open items: Questions requiring client documents or professional judgment.
6. Verification: Accounts tied to the trial balance, calculations recomputed, and source records inspected.
7. Review: Name of the responsible professional and disposition of each proposal.
8. Result: Approved entries posted, rejected proposals, and unresolved items carried forward.

That record supports quality review even if no regulator ever asks for it.

Client-data controls need to follow the account and provider

The IRS guidance tells practitioners to use secure, enterprise-approved AI and not upload sensitive data to unsecured sites.

Before approving a provider, document:

• what data the service receives;
• whether client data is used for model training;
• retention and deletion terms;
• subprocessors and data locations;
• authentication and access controls;
• encryption in transit and at rest;
• incident-response commitments;
• how access is revoked;
• whether the firm can export its activity records.

The firm also needs a rule for consumer accounts. A tool approved under an enterprise agreement may have different privacy and retention terms when used through a personal account.

Consult qualified counsel about the firm’s specific consent, disclosure, and taxpayer-information obligations.

Make professional review visible

“A human is in the loop” is too vague to audit.

For material accounting or tax activity, the record should show:

• what the reviewer received;
• which facts and calculations they checked;
• which proposals they changed or rejected;
• the time of approval;
• what happened after approval.

The review should occur at the point where professional judgment matters—not as a blanket approval of a long agent run after changes have already posted.

AI efficiency also changes the billing conversation

The IRS guidance specifically discusses AI efficiency and billing. It says charging for manual time not actually spent or double billing for AI-assisted tasks may violate Section 10.27 depending on the facts. It also says cost savings should be passed on openly.

Firms should review:

• hourly entries describing AI-assisted tasks;
• fixed-fee engagement economics;
• whether AI vendor charges are passed through;
• how efficiency is reflected in pricing;
• whether staff and clients understand the billing policy.

An activity record helps the firm identify where AI materially contributed. It does not dictate the fee model, but it gives the firm better facts for setting one.

Ten practical steps a firm can take now

• Inventory every AI system that can receive client information.
• Approve providers and account types before staff use them.
• Define prohibited data and procedures.
• Map company and client access to engagement assignments.
• Require verification appropriate to the risk.
• Separate AI findings from proposed professional conclusions.
• Require named approval before accounting changes or advice are finalized.
• Record material reads, proposals, approvals, and outcomes.
• Train staff and test that they follow the procedure.
• Review billing and incident-response implications.

Where Connect can help

Connect gives compatible AI assistants controlled access to QuickBooks Online without placing Intuit credentials in the chat client.

The firm controls workspace and company access. The assistant can gather reports, inspect supporting transactions, and prepare related changes. Connect can require approval before posting and keeps a gateway activity record outside the conversation.

That record does not make a firm compliant. It supplies evidence the firm can use alongside its policies, training, provider review, professional verification, and QuickBooks audit history.

Review security and controls, see Connect for firms, or read the broader QuickBooks AI agent evaluation guide.

This article provides general information, not tax or legal advice. Consult your own advisors about Circular 230, IRC 6713 and 7216, state law, professional standards, and your firm’s specific use of AI.