Direct Answer to the AI Privacy Question
Yes, a tax preparer may be able to use artificial intelligence to assist with parts of a tax return without telling you that a particular generative AI tool was used, depending on the facts, the tool, the preparer’s professional status, and the client agreement. There is not, as of September 29, 2026, a single universal federal rule that says every tax preparer must announce every use of AI. That does not mean disclosure is optional whenever sensitive information is involved. A preparer still must meet applicable privacy, security, confidentiality, tax-practice, and contract duties, and using AI can create risks that would not exist if the work remained inside the firm’s controlled systems.
Also worth reading: Does Your Tax Preparer Need to Disclose When They Use Artificial Intelligence? · What Controls Should an AI Tax Preparer Have Before Filing Returns in 2026? · How Can You Use an AI Budgeting Advisor Without Giving Up Your Financial Privacy?
The practical distinction is between using ordinary software for calculation or document processing and submitting personal tax information to a public generative AI service. A tax program that estimates a deductible automatically is not necessarily the same as uploading a Social Security number, W-2, bank statement, or full return to an external chatbot. The more sensitive the data and the less control the client has over retention, training, access, or deletion, the stronger the case for asking about the tool before information is transmitted.
A simple rule for taxpayers is: ask before you upload, and ask your preparer what they use. A preparer who cannot identify the service, its data settings, or who can access the information may not be able to demonstrate an appropriate level of care. Tax preparation is not a place where privacy should depend on an assumption that “the software is just an assistant.”
How AI Gets Used in Tax Preparation
Tax preparers may use AI for drafting an initial response to a client email, organizing scanned receipts, categorizing ordinary expenses, suggesting questions about a deduction, comparing document dates, or checking a return for inconsistent entries. These uses differ greatly in risk. An internal spell-checker running on an approved platform presents a different question from a public chatbot that may retain prompts, analyze uploaded files, or use information to improve its services.
The key issue is not whether a machine participates in the work. It is what data enters the system, what output is produced, and whether a qualified human reviews the result before it affects a filing. AI can misread handwriting, invent a plausible deduction, misclassify a payment, or create a calculation that appears authoritative but is wrong. Those errors can lead to an incorrect return, amended-return costs, interest, penalties, or questions from the IRS. The taxpayer remains responsible for the information reported on an authorized return even when a preparer or software provider made the error.
The public tax record also does not provide certainty that AI use is always harmless. Tax professionals must consider professional obligations, including the duties of confidentiality, competence, due care, and accurate advice. For some practitioners, Circular 230 duties apply when they represent a taxpayer before the IRS. The Circular 230 regulations do not create a detailed, AI-specific notification procedure in every situation, so the absence of an explicit AI label should not be treated as permission to disregard ordinary privacy protections.
What “Without Telling You” Usually Means
There are several different scenarios hidden in that phrase. A preparer might use a private, firm-approved AI tool for internal research without mentioning it because the client’s data stays within a controlled environment. Another preparer might paste a client’s name and partial tax question into a public chatbot, which could be a different matter even if no SSN or document was included. A third scenario involves a consumer asking an AI service to prepare an entire return, uploading images of every financial record, and assuming the service is secure.
A useful distinction is between disclosure of the tool and consent to disclose the data. Telling a client that AI was used does not automatically make an unauthorized upload acceptable. The client needs enough information to understand what information will be shared, with whom, for what purpose, and for how long. The client may also need to know whether the provider can use the information for model training, whether human reviewers can access it, and whether the firm can delete it.
Written notice is not required for every routine calculation in every tax engagement, but it becomes more important when the preparer proposes to use an external system, share documents with another provider, or rely on automated advice for a material decision. Existing engagement letters, privacy notices, vendor agreements, and state-law requirements can create obligations even when the IRS has not issued an AI-specific rule. A taxpayer should review those documents rather than assume that a verbal assurance is the entire answer.
What to Ask Before a Return Is Filed
Start with a direct question: “Will you or anyone else use generative AI or an external AI service for my return, and will any of my personal documents or data be uploaded?” Follow up with a request for the service name, the administrator of the account, the intended use, and the retention or training setting. If the answer is that an approved internal system is being used, ask what security controls surround it. If an external service is used, ask whether the terms prohibit model training and whether the preparer can delete the uploaded materials.
Taxpayers should also ask who performs the final review. AI-generated text should never be treated as a final tax opinion merely because it sounds confident. A human preparer should compare the return with source documents, resolve missing information, and remain accountable for positions taken. The taxpayer should check names, addresses, filing status, income, deductions, credits, bank information, and any unusual explanation before signing.
Sensitive information should be minimized. A full SSN is usually unnecessary when only the last four digits are sufficient for a routine identity check, and a full document set should not be uploaded when a redacted excerpt can answer the question. Passwords, bank login credentials, and authentication codes should never be placed in a general-purpose chatbot. These measures do not eliminate every risk, but they reduce the amount of information exposed if a provider or account is compromised.
AI Tools Versus Traditional Tax Software
The choice is not simply “AI versus no AI.” Traditional tax software also processes private information, can make errors, and may transmit data to a vendor. The meaningful comparison is between a controlled, purpose-built platform and a general-purpose chatbot used outside the tax firm’s normal workflow.
| Feature | AI-assisted tax service | Traditional tax software or human preparer |
|---|---|---|
| Main advantage | Can help explain, organize, and draft information quickly | Uses established tax rules, forms, and calculation workflows |
| Data control | May involve external prompts, uploads, retention, or provider access | Often uses a defined vendor environment and account permissions |
| Error risk | Can produce invented, outdated, or context-poor answers | Can still have input, formula, or version errors, but errors may be easier to trace |
| Human review | Essential before filing or relying on advice | Expected for advice and generally part of the engagement |
| Typical cost | Can be free to low cost for consumer use, with higher business and privacy-control costs | Individual DIY software may be roughly $0 to $200 per year; professional preparation commonly costs several hundred dollars or more for a simple federal return |
| Best use | Drafting, research organization, and preliminary explanations | Calculations, return preparation, compliance work, and accountable advice |
For most taxpayers, a human preparer using established software and checking the return is the conservative default. AI can be useful as a drafting or research aid when the data is controlled and the output is verified. It should not be the only person deciding whether a deduction is allowable or whether a number belongs on a return.
Common Privacy and Accuracy Mistakes
One common mistake is uploading an entire tax package to ask a simple question. The package may contain identity numbers, account information, medical details, charitable giving, property records, and a complete financial history that exceeds what is needed. Another mistake is treating a fluent response as evidence. Generative AI can produce a confident answer with a wrong threshold, an outdated rule, or an invented source, and it may not distinguish between a general possibility and a taxpayer’s actual eligibility.
Some users assume deleting a chat removes the information everywhere. Retention may occur in backups, logs, administrative systems, or vendor subprocessors, and deletion requests may not cover every copy. Others assume a paid plan is automatically secure. A paid consumer subscription may still permit certain data uses unless the account is configured or the provider’s terms say otherwise. Business and enterprise offerings commonly provide stronger contractual controls, but those controls should be read rather than inferred from the word “enterprise.”
Taxpayers also make the mistake of asking several AI systems to vote on an answer. Agreement among tools does not establish correctness because they may share training data, similar errors, or the same flawed assumption. A better process is to identify the relevant tax year, confirm the rule in authoritative material, compare it with the taxpayer’s documents, and have a qualified professional evaluate exceptions. If a tool cites a source that cannot be opened or verified, the citation should not be trusted merely because it looks formal.
When to Act and When to Avoid AI
Act early in the engagement, not at the last minute. Before sending documents, ask the preparer to explain the privacy policy in writing and identify the workflow. That conversation is particularly important for a business owner, a taxpayer with complex income, a person filing in multiple states, or anyone whose return contains medical, trust, investment, estate, or legal information. These situations can require judgment beyond a generic AI response.
For a low-risk research question, a consumer AI assistant may be acceptable if no identifying information is entered and the answer is independently checked. For drafting an organizer or summarizing a document, a private business tool with a suitable data agreement may be more defensible. For selecting filing status, claiming a credit, reporting cryptocurrency transactions, valuing property, or determining whether a document is required, use authoritative tax guidance and a qualified professional rather than relying on an unreviewed chatbot.
If a preparer cannot provide a clear answer, the taxpayer should pause the upload. That may mean choosing a different provider, using a redaction process, or paying for a human-reviewed engagement. It may also mean declining AI entirely. Privacy and accuracy are worth the extra time, especially where an incorrect return can trigger amended filing fees, interest, penalties, or an audit.
As of September 29, 2026, the safest position is not that AI is forbidden for tax preparers or that every use must be publicly announced. The safer position is that sensitive data should not be sent to an unknown service, AI output should never replace professional verification, and clients should receive enough information to make an informed decision. The relevant law can depend on the jurisdiction, the preparer’s credentials, the data involved, and the service’s terms, so a taxpayer with a specific concern should obtain advice from a qualified privacy or tax professional.