A few years ago, most taxpayers would never have considered sharing a tax notice with a computer program and asking it what to do. Today, that is an everyday temptation. You receive a letter from the IRS, worry about the deadline, and turn to an artificial intelligence tool for a quick explanation. Within seconds, it offers a summary, a checklist, and even a draft response. That convenience is real. So is the risk of giving away information you cannot easily take back.

As an Enrolled Agent representing taxpayers before the IRS and state tax agencies, I believe technology can help people become better informed. But I also believe your tax return, banking records, Social Security number, and financial history deserve far more protection than an ordinary internet question. The important question is not simply whether AI can answer you. It is what happens to the information you provide while asking.

AI can be useful without knowing who you are

AI tools can explain unfamiliar terms, help you prepare questions for a meeting, organize a list of issues, or translate complicated tax language into plain English. For example, asking “What is an IRS Notice of Deficiency?” is very different from uploading a notice containing your full name, address, taxpayer identification number, account details, and proposed balance.

You can ask about a general situation without identifying yourself. Instead of entering “My name is Jane Smith, my Social Security number is …, and the IRS says I owe $38,417,” try: “What should a taxpayer generally review after receiving an IRS notice proposing additional income tax?” The second question can be useful without exposing private information. That distinction is the foundation of safer AI use.

Why the information you enter matters

Different AI products have different privacy settings, retention policies, training practices, enterprise agreements, and security controls. Some conversations may be saved in account history, retained for abuse monitoring, reviewed under specified circumstances, or processed by service providers. A setting that limits model training does not necessarily mean the information is never stored. Deleting a chat also does not necessarily erase every retained copy immediately.

The risk is not limited to a dramatic data breach. You might accidentally share a document through a public link, use a browser extension that can read page content, connect an AI application to cloud storage with broader permissions than intended, or paste a confidential client file into a personal account. You may also receive an inaccurate answer that sounds confident enough to influence an important financial decision.

Before sharing any sensitive material, understand which service you are using, who administers the account, whether content is retained, what third parties can access it, and whether you have a secure alternative. If you cannot answer those questions, do not upload the document.

Seven things I urge taxpayers never to paste into a general-purpose AI chat

First, never enter your online banking username, password, passkey recovery information, or one-time verification code. A legitimate tax explanation does not require those credentials. If a chatbot or someone claiming to assist you asks for a verification code, stop and verify the request through an official channel.

Second, do not enter full bank account or routing numbers, debit card or credit card numbers, expiration dates, or card security codes. These details can be used for unauthorized transactions or impersonation.

Third, protect Social Security numbers, Individual Taxpayer Identification Numbers, dates of birth, driver’s license numbers, and passport details. Identity thieves can combine small pieces of personal data into a convincing profile.

Fourth, avoid uploading unredacted tax returns, W-2s, 1099s, payroll reports, financial statements, or tax transcripts. These documents often contain more identifying information than you notice at first glance.

Fifth, do not paste complete IRS or state tax notices containing your taxpayer identifiers, addresses, barcodes, reference numbers, and account details. Ask about the notice type or share only a carefully redacted excerpt when appropriate.

Sixth, never provide the answers to security questions, recovery codes, or multifactor authentication prompts. Those items are designed to prove your identity, not to improve an AI response.

Seventh, protect information belonging to someone else. If you are a business owner, do not upload employee payroll records, customer data, vendor banking instructions, or a partner’s tax return without a properly authorized and secure process.

A tax notice is not just another PDF

Consider a taxpayer who receives a collection notice and asks AI to draft an appeal. The notice may include a deadline, a specific collection action, and information that changes which rights are available. A generic answer could overlook the difference between a Collection Due Process hearing and another administrative review. Missing a deadline may limit options even when the underlying tax issue is legitimate.

Another taxpayer might ask whether the IRS can levy a bank account and then upload six months of bank statements. Those statements reveal income, expenses, transfers, medical payments, family obligations, and potentially the identities of other people. An AI system does not need all of that information to explain what a levy is.

Tax matters are fact-specific. Filing status, assessment dates, prior notices, appeal rights, business structure, state law, and procedural history can all change the answer. Use AI to learn the vocabulary, not to substitute for a careful review of your actual case.

How to ask better questions while protecting your privacy

Start with the general issue. Ask, “What does an IRS bank levy mean?” or “What documents might an Enrolled Agent request when evaluating an installment agreement?” If you need to describe a scenario, use hypothetical facts and approximate numbers. Replace a real name with “the taxpayer,” a precise balance with “about $25,000,” and an exact employer with “a small business.”

Remove more than the obvious identifiers. Document metadata, filenames, embedded comments, QR codes, account references, and screenshots of browser tabs can expose details. Blacking out text visually may not remove the underlying text layer in a PDF. Use a reliable redaction tool and verify the result before sharing it anywhere. When the issue requires your complete documents, use a secure portal or another method approved by your tax professional rather than a public chatbot.

Also verify any tax law citations the AI provides. An invented court case, outdated IRS form, or inaccurate filing deadline can look persuasive on the screen. Confirm primary authorities through IRS.gov, official state agency websites, and qualified professional advice.

What about AI built into a bank, tax software, or workplace?

A chatbot inside a bank’s authenticated website is not necessarily governed by the same rules as an unrelated public AI service. Likewise, a business AI account may offer administrative controls and contractual protections that a consumer account does not. But the label “secure,” “private,” or “enterprise” is not enough by itself. Review the actual disclosures, permitted uses, retention terms, and access controls.

Use your bank’s official app or website for banking matters. Do not send credentials or verification codes to a third-party assistant, even if it claims to be helping you troubleshoot an account. For tax matters, use your preparer’s or representative’s designated secure exchange method. A trusted professional should be able to explain how confidential documents are collected and handled.

If an AI product offers a feature to connect email, cloud drives, or financial accounts, think carefully before granting permission. Access to an entire folder or mailbox may be much broader than what is needed to answer a single question.

AI-generated tax advice can sound certain and still be wrong

One of the most challenging aspects of AI is its confident tone. It may produce a polished explanation even when it misunderstands a notice, applies the wrong tax year, or misses a state-specific rule. Tax law changes. Agency procedures change. An answer that is generally correct may still be wrong for your facts.

For example, a taxpayer may hear that the IRS has ten years to collect a tax debt and assume the deadline is straightforward. In practice, collection periods can be affected by assessments and events that suspend or extend the period. Or a business owner may ask whether an unpaid payroll tax liability can become personal; the answer may depend on trust fund recovery penalty rules and the person’s responsibility and conduct.

I encourage clients to treat AI-generated answers as a starting point for questions, not as a determination of their rights. If money, deadlines, liens, levies, audits, penalties, or appeals are involved, accuracy matters more than speed.

If you already shared sensitive information

Do not panic, but do act promptly. Stop sharing additional information. Review the AI provider’s privacy controls and available deletion options, recognizing that deletion and retention policies vary. If you exposed a password, change it through the official website or app and enable multifactor authentication. If you exposed a verification code, contact the affected institution through its published contact information.

If account or card numbers were disclosed, contact the bank or card issuer to ask whether monitoring, replacement, or additional safeguards are appropriate. If your Social Security number or tax identity information may be compromised, review the Federal Trade Commission’s identity theft resources and consider an IRS Identity Protection PIN. Monitor account activity and relevant tax correspondence. Business owners should also consider whether any contractual, professional, or legal notification obligations apply.

The response should match the information disclosed. A hypothetical tax question creates a very different concern from a complete tax return with identifying information.

Frequently asked questions

Is it safe to ask ChatGPT or another AI tool a tax question?
It can be reasonable to ask general educational questions without sharing identifiable financial information. Check the product’s privacy settings and independently verify important answers.

Can I upload my tax return to AI for a second opinion?
I generally recommend against uploading an unredacted return to a general-purpose AI service. A full return may contain Social Security numbers, dependent information, bank details, employer information, and other sensitive facts. Seek a secure professional review instead.

Can AI help me respond to an IRS audit letter?
AI may help explain terminology or suggest questions to ask. It should not be relied upon to interpret your specific audit rights, calculate deadlines, or submit a response without careful verification.

What should I remove from a tax notice before using AI?
At minimum, remove names, addresses, taxpayer identification numbers, account numbers, barcodes, reference numbers, and any other details that could identify you or another person. If you cannot confidently redact it, do not upload it.

Does turning off AI training make my information completely private?
Not necessarily. Training settings, chat history, temporary retention, service-provider processing, and legal obligations are separate issues. Read the current policy for the specific service and account type.

Should I trust an AI answer that cites the Internal Revenue Code?
Verify the section, its current applicability, and the underlying facts. A citation can be incomplete, outdated, or fabricated. Official IRS and state sources are better starting points for confirmation.

What is the safest way to send tax records to my representative?
Ask your representative for the firm’s approved secure document-delivery method. Do not assume that a public AI chat, ordinary email attachment, or shared link provides the protections your records deserve.

My advice: protect your information first

I want taxpayers to benefit from useful technology. If an AI explanation helps you understand an unfamiliar IRS term or prepare thoughtful questions before a consultation, that is a positive use. But convenience should never require surrendering your financial privacy.

Here is the simple rule I share with clients: if you would hesitate to post a document publicly, do not casually paste it into an AI chat. Keep credentials and verification codes private. Keep full tax and banking records in secure channels. And when an IRS or state tax problem could affect your wages, bank account, business, or financial future, obtain advice based on the actual documents and applicable law.

At the tax firm of Mike Habib, EA, we help individuals and businesses address IRS and California tax agency matters, including audits, collection issues, and tax resolution concerns. If you have received a notice or need representation, contact our office to discuss the next appropriate step.

Your tax information tells the story of your financial life. Protecting it should always come before getting an instant answer.

Trusted resources

IRS — Identity Protection PIN: https://www.irs.gov/identity-theft-fraud-scams/get-an-identity-protection-pin

IRS — Taxpayer rights: https://www.irs.gov/taxpayer-bill-of-rights

FTC — IdentityTheft.gov: https://www.identitytheft.gov/

FTC — Protecting personal information: https://www.ftc.gov/business-guidance/privacy-security