Can Tax Preparers Use ChatGPT on Client Returns? What 7216 Says

Consider a partner named Marcus, who runs a thirty-person firm. Early in March, a senior named Priya is working through a client's partnership return and cannot make sense of a footnote in a Schedule K-1. She copies the footnote, along with the client's name and the partnership's tax identification number, into a free AI chat window, and asks what it means. The answer is good. It saves her forty minutes. She does the same thing the next day, and so does half her team by the end of the month.

This is an illustration, not a real client. The pattern is real, and it is the reason this question reaches partners. Nobody did anything careless in the ordinary sense. Marcus has simply never decided whether what Priya did is permitted, and if he is asked tomorrow, he will not have a written answer.

This article sets out what the rules actually say, what the IRS has said about AI so far, and what remains unsettled. It is general information, not legal advice, and a question this consequential deserves a review by your own counsel.

Here is what matters most:

• Section 7216 of the Internal Revenue Code makes it a crime for a tax return preparer to knowingly or recklessly disclose or use tax return information outside the narrow permitted purposes.
• The regulation defines "disclosure" as making tax return information known to any person in any manner whatever. A client's name and identifying number, furnished in connection with a return, count as tax return information.
• On June 24, 2026, the IRS Office of Professional Responsibility issued Alert 2026-19, its first formal guidance on AI in tax practice. It did not prohibit AI, and it did not settle whether a given tool counts as a permitted recipient.
• The unresolved question is whether sending return information to an AI tool is a disclosure that needs the client's written consent. Credible practitioners disagree, so the careful course is to treat it as a disclosure until counsel tells you otherwise.
• The firms best placed are those that decided in advance which tools staff may use, on what terms, and with what client consent.

What Section 7216 actually restricts

Start with the text, because most of what circulates online paraphrases it loosely.

Section 7216 and its regulation, Treasury Regulation section 301.7216-1, apply to a "tax return preparer", defined as any person engaged in the business of preparing or assisting in preparing tax returns. That covers the firm, its partners and its staff.

The offence is a knowing or reckless disclosure or use of tax return information. Paragraph (a) of Treasury Regulation section 301.7216-1 describes a violation as a misdemeanor carrying up to one year of imprisonment and a fine of not more than $1,000, together with the costs of prosecution. A separate civil penalty exists under section 6713. The IRS alert discussed below refers to both.

Two definitions in the same regulation do most of the work for anyone thinking about AI.

"Disclosure" means the act of making tax return information known to any person in any manner whatever. There is no carve-out for the manner of transmission, and no exception for a quick question.

"Tax return information" means any information, including a taxpayer's name, address or identifying number, furnished for or in connection with the preparation of a return. That is a wide definition. A prompt does not need to contain a full return to qualify. A name and an identifying number may be enough.

The regulation does allow disclosure without consent in defined circumstances. Two are worth knowing for this discussion. A preparer in the United States may disclose to another preparer for auxiliary services, but only where the services are not substantive determinations or advice affecting the tax liability. And a preparer may disclose to a contractor for tasks such as programming, maintenance, repair and testing, but only as needed for those services, and the preparer must give the recipient written notice of the requirements and penalties of sections 6713 and 7216. The regulation also draws a separate line at recipients located outside the United States, which is why where a vendor operates matters.

Everything else generally needs the client's written consent, given before the disclosure, in the form the regulation requires.

What the IRS has said about AI so far

The IRS Office of Professional Responsibility published Alert 2026-19, "Introductory Guidelines for Responsible AI Use in Federal Tax Practice", on June 24, 2026. It is worth reading in full because it is short and because it is the nearest thing to an official position.

It does not ban AI. It describes generative AI as holding promise to improve efficiency and professional services in tax practice. It then ties AI use to duties practitioners already have under Circular 230:

• Due diligence under section 10.22: the alert says due diligence requires verifying the accuracy of facts, citations and calculations produced by AI, and that practitioners cannot rely solely on AI.
• Competence under section 10.35: practitioners must understand both the law and the technology they use for clients.
• Fees under section 10.27: cost savings should be passed on openly, with billing that reflects the efficiencies gained.
• Firm procedures under section 10.36: firms must have internal policies and procedures for compliance in the AI space.

On client data, it is direct. It says practitioners must handle all client data using only secure, enterprise-approved AI, and that they should never upload sensitive data to unsecured sites. It also states that civil and criminal preparer penalties apply for unauthorized use or disclosure of tax return information.

What the alert does not do is answer the specific question Marcus has. It does not say that sending return information to a particular type of AI tool is, or is not, a disclosure under section 7216. It does not say whether an enterprise agreement changes the analysis. It points at the existing rules and tells practitioners to comply.

What is settled and what is not

Separating the two is more useful than any single answer.

Reasonably settled

• Return information is protected by section 7216 and section 6713, and the penalties are real.
• Disclosure is defined broadly, and identifying details alone can be tax return information.
• Written client consent, obtained before disclosure, is the ordinary route where no exception applies.
• The IRS expects firms to have their own written policies on AI, and expects practitioners to verify what AI produces.
• The Federal Trade Commission's Safeguards Rule requires a written security program, including oversight of the service providers who handle customer information (16 CFR 314.4, including paragraph (f)). The IRS tells tax preparers that failing to maintain such a plan can lead to an FTC investigation. The FTC has been reconsidering parts of the rule, so check its current text.

Not settled

• Whether an AI provider receiving a prompt is merely processing information or making a substantive determination, which matters because of the auxiliary services exception. A panel at the American Bar Association Tax Section's virtual fall meeting, dated October 6, 2026, was framed around exactly this question, which tells you it was still open in the fall.
• Whether an enterprise agreement with an AI vendor places the vendor within an exception, or still requires consent.
• Whether removing a name and number from a prompt takes the remaining content outside the definition. Figures derived from a client's return may themselves be tax return information.
• How specifically a consent form must name the AI tool and its operator.

You will find confident answers to all four online, usually on a vendor's blog. Be wary of them. A firm selling consent templates or an AI product has a commercial reason to sound certain. The honest position, as of today, is that these four points lack authoritative resolution.

A workable approach for a firm of about thirty

Firms in the thirty-person range tend to share one worry: their people are already using these tools, the partners do not control which, and the partners are the ones who answer for it. That is a reasonable worry, and it is the right one to design around. The aim is not to stop your staff using AI. It is to give them a safe, approved way to do it, so that review and judgment stay with your people and the machine handles the reading, summarising and first-draft work.

In our experience, a sensible order of decisions looks like this.

1. Decide what may never go into a general-purpose chat tool. Names, identifying numbers, addresses, account numbers and any document taken from a client file should be on the list. Put it in writing, in one page, and put it in front of every person who touches a return.

2. Name the approved tools, and nothing else. The alert's phrase is "secure, enterprise-approved". Make that concrete: a short list, each tool reviewed by someone senior before it goes on it. Read the vendor's current terms on retention and on whether your data is used to improve their product, because those terms differ by plan and change over time.

3. Treat the AI vendor like any other service provider. The Safeguards Rule requires reasonable steps to select and retain service providers capable of maintaining appropriate safeguards, a contract that requires those safeguards, and periodic reassessment. An AI tool that touches client information belongs in that file next to your hosting provider.

4. Where the law is unclear, use consent. Written consent, obtained before the data goes in, naming the tool, is the conservative answer to the unsettled questions. It costs a line in the engagement letter and a signature. It does not cure a disclosure that has already happened, so the order matters.

5. Keep a record. Which tool, which client, which return, who approved it. If a regulator or an insurer ever asks, a log answers in minutes what memory cannot.

6. Keep the human review. The alert is explicit that practitioners cannot rely solely on AI. Whatever a tool produces, a qualified person reviews and signs for it. This is also the best answer to staff who worry that AI is there to replace them. It is not, and under the IRS's own framing it cannot be.

Where your data sits is part of the answer

The firms we speak to give one reason more than any other for hesitating: they cannot say where client information goes once it leaves their systems. That is a fair concern and it is answerable.

A system built for a firm can run inside the firm's own cloud account, on infrastructure the firm controls, so that client documents are read and processed without being handed to a public chat service. Access can be limited by role, every action can be logged, and the firm decides how long anything is kept. None of that removes the need for the policy, the contract and, where counsel advises it, the consent. It does make each of those questions much easier to answer, because the data never left your boundary in the first place.

For the wider question of how a vendor should fit around the software you already run, see our piece on the real test for an AI vendor pitch, and for what firms should expect from AI in return preparation, AI tax preparation software. This sits within the broader work of automating tax documents in a CPA firm.

FAQ

Is it illegal for a tax preparer to use ChatGPT?

Not as such. The IRS alert treats AI use as conditionally acceptable and sets duties around it. What can be unlawful is a knowing or reckless disclosure or use of tax return information that no exception covers and the client has not consented to. Whether a particular tool counts as a disclosure is the unsettled part.

Does removing the client's name make it safe?

It lowers your exposure, but it may not remove it. The regulation defines tax return information broadly, and figures derived from a client's return can still qualify. Treat redaction as a helpful precaution, not as a reason to skip the policy or the consent question.

Does a paid enterprise AI plan change the answer?

It changes the vendor's data terms, such as retention and use of your content, which matters for the Safeguards Rule and for the IRS's "enterprise-approved" language. Whether it places the vendor within a section 7216 exception, or leaves consent necessary, is one of the open points. Ask your counsel, and until then lean on consent.

Do we need client consent before using AI at all?

Consent is the standard route for any disclosure that no exception covers. Because the position on AI tools is unsettled, many firms choose to obtain written consent, naming the tool, before client information is entered. Have your counsel review the wording, and obtain it before the data goes in, since later consent does not cure an earlier disclosure.

Want to see how this works inside your own systems?

If you would like to see how client documents can be read and organised inside your own cloud, with your people reviewing the output, book a free strategy call. We will walk through your current tools and show you what stays within your control.

Read next: AI Automation for CPA & Accounting Firms