How to Handle a Tax Document Upload to an Unapproved AI Tool

What should a firm do first when a tax document goes into an unapproved AI tool?

Stop the spread and preserve the facts. Have the employee stop using the tool, capture what was uploaded and when, and report it to the firm's Qualified Individual the same day, before anyone deletes the chat. Treat it as a possible tax preparer data breach until the facts show otherwise.

The instinct is to delete the conversation and move on. Deleting first destroys the evidence the firm needs to decide whether anyone has to be notified. Capture the record, then contain it.

The FTC Safeguards Rule treats a tax preparation business as a financial institution and requires a written incident response plan and a designated Qualified Individual. The IRS has said in News Release IR-2026-92 that federal law requires tax and accounting professionals to maintain a written information security plan. An AI upload is the kind of event that plan should already cover. If it does not, this is the time to write it down.

What facts does the firm need to capture?

Enough to answer three questions: what data left, where it went, and who can now see it. Write the answers down while the employee's memory is fresh.

• The tool and account: product name, web address, and whether the employee used a personal account, a free tier, or a business plan.
• The time: when the upload happened and when the firm learned of it.
• The content: which documents or text, which clients, and which data elements, such as names, Social Security numbers, bank account numbers, income figures or return copies.
• The count: how many individuals are involved, including spouses and dependents listed on a return.
• The vendor terms: what the tool's terms say about retention, model training and human review. OpenAI, for example, states that it may train on content from individual accounts unless the user opts out, and does not train on business plans by default.

Screenshots or an export of the conversation, saved to a restricted incident folder, are usually enough. Then delete the chat and any uploaded files from the tool, and change the employee's password if the account was a firm account.

Is an upload to an AI tool a Section 7216 disclosure?

An upload to an AI tool can be a Section 7216 disclosure. Treasury regulations define disclosure of tax return information broadly, and they do not exempt AI tools. Whether a specific upload was an improper disclosure depends on the facts and is a question for the firm's counsel.

26 U.S.C. 7216 makes it a misdemeanor for a preparer to knowingly or recklessly disclose or use tax return information other than to prepare a return, and section 6713 adds a civil penalty of $250 per improper disclosure, capped at $10,000 per calendar year. The regulations define disclosure as making tax return information known to any person in any manner whatever.

The regulations permit some disclosures without client consent, including to a contractor for software services, to the extent necessary, with written notice of the 6713 and 7216 requirements. An unapproved tool, used without that notice and without a contract, is hard to fit into those exceptions. Other disclosures require a signed, dated client consent.

When does an AI upload become a reportable tax preparer data breach?

An AI upload becomes reportable when the facts show unencrypted customer information was acquired without authorization, which can trigger FTC and state notice rules. The thresholds and timing differ, so the firm needs the count and data elements before it decides.

• FTC: the Safeguards Rule defines a notification event as acquisition of unencrypted customer information without authorization. If it involves at least 500 consumers, the firm must notify the FTC no later than 30 days after discovery.
• California: the state breach statute requires a business to notify California residents whose unencrypted personal information was, or is reasonably believed to have been, acquired by an unauthorized person, and to send a sample notice to the Attorney General when more than 500 residents are notified from a single breach.
• IRS: if the review concludes client data was compromised, the IRS data theft guidance directs tax professionals to contact their local Stakeholder Liaison so the IRS can watch for fraudulent returns, and to notify state tax agencies.

Whether an AI vendor receiving data through its normal service counts as unauthorized acquisition is a judgment call, and reasonable people will reach different answers on different facts. Document the analysis, the people involved and the conclusion either way. That record matters as much as the decision.

What does the firm tell the client?

Tell them what the review found, once counsel agrees on what any notice must say. A plain account of what happened and what the firm changed protects the relationship.

Where no formal notice is required, the firm can still choose to contact the affected client directly. Keep the message factual: what was uploaded, which tool, what the firm did to remove it, and what the firm changed so it does not happen again. Avoid guessing about what the vendor did with the data. For the wider conversation, see how to answer a client who asks whether you use AI on their return.

How do you keep it from happening again?

Find out why the employee reached for the tool, then close that gap with an approved option and a technical control. Pair any policy reminder with a control that works during busy season.

• Give staff an approved AI tool with business terms, and say plainly which client data it may handle.
• Block unapproved AI sites on firm devices, and use browser DLP rules that recognize Social Security numbers and bank account numbers.
• Add an AI section to the WISP, using IRS Publication 5708 as the starting template.
• Update security training to include AI uploads, which the Safeguards Rule requires firms to keep current with identified risks.
• Add AI uploads to the incident response plan as a named scenario with an owner and a checklist.

What should the firm do this quarter?

Rehearse the response before filing season. A short tabletop exercise with the Qualified Individual, a partner and the IT provider shows where the plan is thin.

• Write a one-page AI upload checklist based on this article and file it with the incident response plan.
• Confirm who the firm's IRS Stakeholder Liaison contact is and where the FTC reporting form lives.
• Inventory which AI tools staff use today, approved or not, using sign-in and spend data.
• Turn on DLP for tax identifiers in email and browsers.

How does s90 help?

s90 monitors AI use for CPA firms and tax practices through prompt monitoring and incident response, so an upload to an unapproved tool is flagged when it happens. Our agreements commit to notifying you of a breach within 72 hours and producing evidence within 24 hours, which gives your Qualified Individual the facts needed to make notice decisions. If you want to test your response before filing season, we can run the tabletop with you.