What client data can a tax practice put into an AI tool?

A tax practice now runs on two kinds of software: the tax software it has licensed for years, and a growing list of AI tools staff want to use for research, letters and workpapers. For a managing partner, the hard question about AI in tax preparation is which client data may go into which tool, and who decides.

What counts as client data in a tax practice?

Almost everything a client gives you, and much of what you create from it. The IRS describes tax return information as all the information a preparer obtains from a taxpayer, including the computations, worksheets and printouts the preparer creates (IRS Section 7216 FAQs; 26 CFR 301.7216-1).

That definition is broader than most staff assume. A spreadsheet of depreciation schedules, a summary of a client's K-1s, or an email recapping a planning call can all be tax return information. If it came from the client or was built from what the client provided, start from the assumption that it is covered.

Which rules govern client data and AI in tax preparation?

Four sets of rules overlap: Section 7216, Circular 230, the FTC Safeguards Rule and, for CPA firms, the AICPA Code of Professional Conduct. None of them bans AI. Each limits who receives client data and under what terms.

• Section 7216 and Section 6713. A knowing or reckless unauthorized disclosure is a crime punishable by up to one year in prison and a $1,000 fine, and Section 6713 adds a civil penalty of $250 per improper disclosure or use (IRS Section 7216 FAQs). The regulations allow some disclosures without consent, such as to another preparer in the United States who assists with the return, and treat contractors who receive return information as preparers bound by the same rules (26 CFR 301.7216-2).
• Circular 230. In June 2026 the IRS Office of Professional Responsibility issued Issue Number 2026-19, introductory guidelines on AI in federal tax practice. It says practitioners "must strictly handle all client data using only secure, enterprise-approved AI," tells them to "never upload sensitive data to unsecured sites," and notes that Section 10.51(a)(15) of Circular 230 prohibits willful disclosure or use of tax return information.
• FTC Safeguards Rule. Tax preparers are financial institutions under the rule (16 CFR 314.2). 16 CFR 314.4 requires a firm to identify and manage its data and systems according to their importance, to control access, and to oversee service providers through selection, contract terms and periodic assessment. An AI vendor that receives customer information is a service provider.
• AICPA Code. ET section 1.700.040 addresses disclosing confidential client information to a third-party service provider: the member needs a contractual agreement that the provider will maintain confidentiality, or the client's specific consent (AICPA Code of Professional Conduct).

What client data can go into an AI tool?

It depends on the data and the tool, so the practical answer is a short data map. Sort information into tiers, then match each tier to the tools approved for it. The tiers below are a common way to organize the decision. Your own counsel should confirm where the lines fall for your firm, using the review list later in this article.

Tier 1: no client data

Public law and guidance, IRS publications, firm templates, marketing copy and fact patterns written from scratch. This tier can go into any AI tool the firm has approved for business use, including general research tools. The Circular 230 duties to check accuracy still apply to the output.

Tier 2: client facts with identifiers removed

A planning question rewritten without names, Social Security numbers, EINs, addresses, account numbers or unusual details that point to one client. This lowers the risk that anyone could be identified, but it does not change the legal status of the information (see the next section). Keep this tier in firm-managed AI accounts, not personal ones.

Tier 3: identified return information

Source documents, drafts, workpapers and anything with a name or taxpayer identification number. This tier belongs only in tools the firm has vetted and approved under contract: an enterprise AI account or an AI feature inside licensed tax or practice software, with terms covering confidentiality, retention, model training and data location. Contractors that receive return information must get written notice of the Section 7216 rules (IRS Section 7216 FAQs). When a use falls outside the regulatory exceptions, the regulations require the taxpayer's signed consent first, in the format set by Rev. Proc. 2013-14 for Form 1040 filers.

Tier 4: never

Any client data in a free or personal AI account, a browser extension nobody approved, or a site with no contract behind it. The OPR guidelines name unsecured sites directly, and the Safeguards Rule's service provider oversight cannot be met for a vendor the firm has no agreement with. For the staff-facing side of this rule, see how to keep client tax documents out of public AI.

Does removing client names make data safe for AI?

Not by itself. Removing names and numbers lowers the chance of identifying a client, but the information was still obtained in connection with preparing a return and can remain tax return information.

The only anonymity standard in the Section 7216 regulations sits in a narrow exception for statistical compilations. A compilation counts as anonymous only if it cannot be associated with a particular taxpayer, directly or indirectly, and it may not disclose an aggregate figure drawn from fewer than ten returns (26 CFR 301.7216-2). A rewritten planning question about one client does not fit that exception. Treat Tier 2 as risk reduction and keep it inside approved tools.

What changes when an AI vendor processes data outside the United States?

Stricter consent rules may apply. The IRS says that when return information goes to a preparer located outside the United States, the taxpayer must sign a consent, and Social Security numbers must generally be redacted or masked first unless specific consent and safeguards conditions are met (IRS Section 7216 FAQs).

Many AI services process data in several regions by default. Before approving a Tier 3 tool, find out where prompts, files and logs are stored and processed, whether a United States data residency option exists, and whether subprocessors outside the country can see client content. Then ask counsel whether the offshore consent rules apply to that setup.

What should a tax practice do this quarter?

Write the data map down, attach approved tools to each tier, and make it enforceable before the 2027 filing season peaks.

• List every AI tool in use, including AI features switched on inside software you already license.
• Assign each tool a highest permitted tier and record it in your WISP. IRS Publication 5708 provides a WISP template you can extend with an AI section.
• For each Tier 3 tool, file the contract terms on confidentiality, retention, training and data location, plus the written Section 7216 notice.
• Decide with counsel which Tier 3 uses need taxpayer consent, and update engagement letters and consent forms to match.
• Block personal AI sign-ups on firm devices and add data loss prevention rules for Social Security numbers, EINs and bank account numbers.
• Train staff on the four tiers with real examples from your practice, and keep the attendance record.

What should your counsel review before staff use AI with client data?

Have your own counsel review four decisions before any Tier 3 tool goes live, because each one turns on your contracts and your vendor's setup. Schedule the review before filing season, and again whenever you add a tool or a vendor changes its terms.

• Exception or consent. Which Tier 3 tools fit a Section 7216 exception and which need the taxpayer's signed consent (26 CFR 301.7216-2 and 301.7216-3). Bring the vendor contract and a plain description of what the tool does with client data.
• Consent forms and engagement letters. Whether your consent forms meet Rev. Proc. 2013-14 and whether your engagement letters describe AI use accurately. Bring the current forms and letters.
• Processing outside the United States. Whether the vendor's data locations or subprocessors trigger the offshore consent and Social Security number masking rules. Bring the vendor's data residency terms and subprocessor list.
• Vendor contract terms. Whether the confidentiality, retention and model training terms meet AICPA ET section 1.700.040 and the service provider requirements in 16 CFR 314.4(f). Bring the contract, data processing addendum and security documentation.

How s90 helps

s90 helps tax and accounting firms turn an AI policy into controls: an inventory of the AI tools staff actually use, firm-managed accounts for approved tools, data loss prevention that stops identifiers from reaching unapproved sites, and WISP documentation that matches what the systems enforce. If you want a second set of eyes on your data map before filing season, we are glad to talk.