Can accountants use ChatGPT? What section 7216 says about client tax data

The Morse teamPublished October 25, 202610 min read

Yes, accountants can use ChatGPT, but not with a client's return information unless a regulation allows the disclosure or the client signed a consent first. As of October 2026, the Treasury regulations under section 7216 of the Internal Revenue Code define a disclosure as "the act of making tax return information known to any person in any manner whatever," and typing a client's figures into an AI chat makes them known to the company that runs it.

This isn't tax or legal advice. It's about where client data goes when you use AI, and what the rules ask of you before it goes.

On October 25, 2026, we read the statute, the Treasury regulations, the FTC Safeguards Rule, the IRS's security publications and its June 2026 AI guidelines, the AICPA code, and OpenAI's, Anthropic's, and Google's own pages on training and retention. The AI companies deserve some credit. Their business plans don't train on your data by default (OpenAI, Anthropic, Google), and OpenAI lets eligible Enterprise customers keep both storage and model processing in the United States.

Is pasting a client's return into ChatGPT a disclosure under section 7216?

Yes, by the regulation's own definitions. Tax return information is "any information, including, but not limited to, a taxpayer's name, address, or identifying number, which is furnished in any form or manner for, or in connection with, the preparation of a tax return," plus anything you "derive or generate" from it (26 CFR 301.7216-1). The rule covers income tax returns under chapter 1 of the Code.

Two penalties attach. Section 7216 makes a knowing or reckless disclosure a misdemeanor, with a fine of up to $1,000 ($100,000 in identity theft cases) and up to a year in prison (26 U.S.C. 7216). Section 6713 adds a civil penalty of "$250 for each such disclosure or use," capped at $10,000 a calendar year, and its text has no knowing-or-reckless test (26 U.S.C. 6713).

Does the auxiliary services exception cover AI tools?

Not for the work most accountants want AI to do. You may disclose to another tax return preparer without consent only if that preparer is "located in the United States," and only "so long as the services provided are not substantive determinations or advice affecting the tax liability." The regulation defines a substantive determination as "an analysis, interpretation, or application of the law" (26 CFR 301.7216-2(d)(1)). Its Example 4 is a preparer who forwards a client's information to another firm to ask whether a business expense is deductible. Consent is required "because B's services involve a substantive determination." Asking a chatbot the same question about a client's expense works the same way.

The second route covers contractors working on "equipment or software used for purposes of tax return preparation," only "to the extent necessary," and only if every individual who receives the information gets "a written notice" of the penalties under sections 6713 and 7216. Example 2 calls a disclosure to a contractor's employee who never got that notice "an impermissible disclosure." Signing up for a chatbot sends no such notice to anyone. Location counts on its own, and Example 3 requires consent when a contractor abroad only views return data stored on a server in the United States.

Does removing names and Social Security numbers make it safe?

It lowers the risk, but the figures stay tax return information. The definition isn't limited to identifiers, and it includes "statistical compilations of tax return information, even in a form that cannot be associated with, or otherwise identify, directly or indirectly, a particular taxpayer" (301.7216-1(b)(3)(i)(B)). Jupid's guide, a top result for this question, says that once identifiers are stripped, what's left is no longer tax return information and "you can use it freely" (Jupid). The regulations allow even anonymous compilations only for the limited purposes in 301.7216-2(o).

Social Security numbers get a stricter rule. For Form 1040 filers, a preparer in the United States "may not obtain consent" to disclose the number to a preparer outside the United States, and must "redact or otherwise mask" it first (26 CFR 301.7216-3(b)(4)). Take out SSNs, EINs, names, addresses, and account numbers before any AI tool sees a figure, and find out where that tool processes what you send.

Can a client consent to you using AI on their return?

Yes, in writing, signed and dated, before the first disclosure. The consent must name you, the client, the purpose, "the specific recipient (or recipients)," and the information to be disclosed. Making your services depend on it makes it involuntary, with a narrow exception for sharing with another preparer who helps with the return (301.7216-3(a)). For 1040 filers, Revenue Procedure 2013-14 prescribes statements in a fixed order, including "If you consent to the disclosure of your tax return information, Federal law may not protect your tax return information from further use or distribution." A consent with no stated term is "valid for one year from the date of signature" (IRS).

Once the information reaches an AI company that isn't itself bound by section 7216, that company's own terms decide what happens to it.

What does the FTC Safeguards Rule require before you use an AI tool?

It requires you to treat the AI company as a service provider. The rule says "an accountant or other tax preparation service that is in the business of completing income tax returns is a financial institution," and a service provider is anyone that "receives, maintains, processes, or otherwise is permitted access to customer information" through services to you (16 CFR Part 314). You have to choose providers "capable of maintaining appropriate safeguards" and require those safeguards "by contract." You must encrypt customer information "both in transit over external networks and at rest." Since May 13, 2024, a breach of unencrypted information about 500 or more consumers must be reported to the FTC "no later than 30 days after discovery."

The IRS turns that into the written information security plan, or WISP. Publication 5708 (Rev. 8-2024) says "the law requires you to have one," and lists "Online (Web-based) applications, portals, and cloud software applications" among the places client data may be stored (IRS). Publication 4557 (Rev. 6-2024) warns that skipping one "may result in an FTC investigation" (IRS). Line 11 of the PTIN renewal form, Form W-12 (Rev. October 2026), asks you to confirm you know preparers "are required by law to create and maintain a written information security plan" (IRS). Any AI tool your staff paste client data into belongs in that plan.

What does the IRS say about using AI in tax practice?

Use only secure tools your firm has approved. On June 24, 2026, the Office of Professional Responsibility issued "Introductory Guidelines for Responsible AI Use in Federal Tax Practice," Issue Number 2026-19. It flags the risk of disclosure "especially when data is uploaded to unsecured or public systems" and says "Practitioners must strictly handle all client data using only secure, enterprise-approved AI" (IRS Office of Professional Responsibility). It also warns of client data "repurposed by the program to respond to an inquiry concerning another client." Circular 230 counts "willfully disclosing or otherwise using a tax return or tax return information in a manner not authorized by the Internal Revenue Code" as disreputable conduct (31 CFR 10.51).

Do business AI plans satisfy section 7216 or the AICPA code?

No. As of October 2026, the business terms for ChatGPT, Claude, and Gemini rule out training on your data by default (OpenAI, Anthropic, Google), which settles the training question and none of the others above. OpenAI also lets eligible new Enterprise and Edu customers keep content stored, and inference run, in the United States (OpenAI), which speaks to the regulations' location test.

No business contract is a section 7216 consent, and none turns "is this deductible?" into something other than a substantive determination. The AICPA code adds its own test. Before disclosing confidential client information to a third-party service provider, a member should either have "a contractual agreement" to keep it confidential, with "reasonable assurance" about the provider's procedures, or "obtain specific consent from the client" (interpretation 1.700.040, in the code as updated through September 2026, AICPA). The code's carve-out for "software application hosting" sits in a different interpretation, 1.150.040, and isn't in this one.

What does each rule mean when you paste client data into AI?

We read each rule's own text on October 25, 2026. The right-hand column applies it to a prompt with client data in it.

How Morse keeps client tax data off AI company servers

Every rule in the table is about the same exposure. Once you press send, your client's figures sit, readable, with an AI vendor whose contract you didn't negotiate. Whether a disclosure is allowed, and whether you need consent, stays with the rules above. Morse is built so fewer readable copies exist. In E2EE, only a verified enclave can read your message, and Morse's servers relay ciphertext they can't read. In every mode, your chat history stays in your browser, not on our servers.

There's no training default to find, because Morse doesn't train AI on your content. Morse's servers don't keep messages or log message content.

• E2EE, for a client's figures, once the consent question is settled. Your browser checks the secure enclave's hardware attestation, then encrypts the message so only that enclave can read it. The reply is decrypted only in your browser. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed. E2EE runs open-weight models like GLM, Kimi, and DeepSeek without web search, so look up the code section in a separate chat. Strip names and SSNs here too, since a question about the figures rarely needs them.
• Private, for tax research and templates with no client in them. Private is the default for a new chat. Open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok, run on providers that agreed to keep nothing, because Morse has zero data retention agreements with its inference, search, and other providers. Use it to work through a regulation or draft a template (what zero retention means).
• Anonymous, for drafting with GPT, Claude, or Gemini. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider sees the text without knowing it's you and may keep it for a limited time to check for abuse, under its own policy, so keep client names, SSNs, EINs, and return figures out of it.

In every mode, Morse doesn't sell your data or show ads. Your history is stored in your browser, encrypted, and unlocking it takes two keys, one that never leaves your device and one our server gives you when you sign in. An attached file is sent with the message and never stored by Morse. Because we don't store chats, we usually have very little to give authorities (privacy policy). Clearing the browser, losing the device, or deleting your account deletes your chats, and Morse can't recover them, so keep anything that belongs in the client file in your own records system.

Try Morse from the home page without an account, or start free with no card needed. E2EE, the mode for client figures, comes with Pro at $19 a month.