How to Use AI to Review Engagement Workpapers Safely

Can a CPA firm use AI to review engagement workpapers?

Yes, with conditions. No AICPA rule bans AI tools. The confidentiality rule, the federal tax disclosure rules and the quality management standards apply to an AI tool the same way they apply to any outside party that touches client files. The safe version runs on a firm-managed AI account under a business contract, keeps client data inside that account, and puts a named reviewer behind every output.

Workpaper review is one of the most practical uses of AI for CPA firms. Tools can tie out schedules, flag unusual variances, summarize lease and debt agreements, check that tickmarks are explained and draft first-pass review notes. The risk sits in what gets uploaded: trial balances, payroll registers, bank statements, tax return data and client correspondence. Once that material lands in a tool the firm does not control, the firm cannot pull it back or prove who saw it.

Which rules apply when workpapers go into an AI tool?

For most firms, three: the AICPA confidentiality rule, Section 7216 for tax return information, and the FTC Safeguards Rule for firms that prepare returns.

• AICPA confidentiality. Interpretation 1.700.040 of the AICPA Code of Professional Conduct treats disclosing confidential client information to a third-party service provider as a threat to compliance with the Confidential Client Information Rule. The member either contracts with the provider to keep the information confidential and obtains reasonable assurance that it has appropriate procedures, or obtains specific client consent before disclosing. An AI vendor that receives workpapers is a third-party service provider.
• Section 7216. When workpapers contain tax return information, 26 U.S.C. 7216 and its regulations apply. Knowingly or recklessly disclosing or using that information for a purpose other than preparing a return is a misdemeanor punishable by up to one year in prison and a fine of up to $1,000. Section 6713 adds a civil penalty of $250 per improper disclosure or use. The regulations define disclosure as making tax return information known to any person "in any manner whatever."
• FTC Safeguards Rule. An accountant or tax preparation service in the business of completing income tax returns is a financial institution under 16 CFR 314.2(h)(2)(viii). The rule's service provider requirements in 16 CFR 314.4(f) call for selecting providers capable of maintaining appropriate safeguards, requiring those safeguards by contract and periodically assessing them. Our FTC Safeguards Rule checklist for tax preparers covers the rest of the rule.

Does SQMS No. 1 apply to AI tools used on engagements?

Yes. SQMS No. 1 makes an AI tool used on engagements a firm resource that has to be managed and documented. SQMS No. 1 includes a quality objective, in paragraph 33(f), that "appropriate technological resources are obtained or developed, implemented, maintained, and used" to operate the firm's system of quality management and perform engagements.

Firms had to have their systems of quality management in place by December 15, 2025. The AICPA says firms must evaluate the system within one year of implementation, or by December 15, 2026, and annually after that. If engagement teams use AI, the first evaluation is a natural place to show why the firm chose its tool, how it is configured, who may use it and how outputs are checked. A tool someone signed up for with a personal email address will be hard to defend in that review.

What does a safe AI workpaper review setup look like?

One approved tool, a business contract, client data that stays in the firm's environment, and a person who owns every conclusion. Use this checklist.

• Pick one business-tier AI tool and confirm it does not train on firm data. Consumer plans often allow training by default. OpenAI, for example, says it may use content from Free, Plus and Pro accounts to train its models unless the user opts out, while its business plans are excluded by default.
• Get the contract right. The agreement should cover confidentiality, data retention and deletion, subprocessors, breach notice and data location. This is the contract interpretation 1.700.040 and 16 CFR 314.4(f) point toward.
• Keep AI work inside the engagement. Use a separate project or workspace per client, or an AI feature built into your audit platform, so prompts and outputs stay with the engagement and are deleted on the same schedule.
• Block the alternatives. Use web filtering to block unapproved AI sites on firm devices, and data loss prevention (DLP) rules to stop Social Security numbers, EINs and bank account numbers from being pasted or uploaded anywhere else.
• Define approved tasks. Start with tie-outs, variance flags, agreement summaries and first-draft review notes. Add tasks after the team has a track record.
• Require human review. The team member who runs the prompt checks every output against source documents before it reaches a workpaper. The reviewer of record stays a person.
• Keep a record. Retain the prompts and outputs that affected the work, or a short note of how AI was used, in the engagement file.

When does a firm need client consent?

A CPA firm needs specific client consent when it shares confidential client information with an AI vendor without a confidentiality contract in place, and Section 7216 can require written taxpayer consent before tax return information is used or disclosed outside return preparation. Under interpretation 1.700.040, specific client consent is the alternative to a confidentiality contract with the provider. The Section 7216 regulations set separate exceptions and a written consent process. Whether a specific AI vendor and workflow fits one of those exceptions is a question to settle with counsel before tax data goes into the tool.

Many firms also add a plain-language AI paragraph to engagement letters. That helps with client expectations. It does not replace the contract or any consent the rules require.

What should a CPA firm do this quarter?

Get the controls in place before busy season and before the first SQMS No. 1 evaluation.

• Inventory the AI tools staff already use, including browser extensions and AI features inside audit, tax and practice management software.
• Choose one approved tool and sign the business agreement.
• Turn on web filtering and DLP for unapproved AI sites.
• Write a one-page AI procedure covering approved tasks, review and documentation, and reference it in your WISP, which the IRS reminds tax professionals federal law requires. Our guide to IRS WISP requirements shows what the plan needs before the 2027 filing season.
• Train engagement teams, then spot-check a few engagement files in January.
• Include AI tools in the SQMS No. 1 evaluation due by December 15, 2026.

How s90 helps

s90 sets up AI guardrails for CPA firms: approved AI tools on firm-managed accounts, web filtering for unapproved AI apps, DLP rules for client identifiers and alerts when new AI apps appear. We document the controls so they line up with your WISP and your quality management records. If you want a second look at how your engagement teams use AI today, we are happy to talk it through.