The legal risks of AI tax advice fall on two people at once: the filer, who signs the return and owes whatever the IRS later says is owed, and the company whose software produced the numbers, which can face preparer penalties, unauthorized-practice claims, privacy enforcement, and FTC action over deceptive marketing. The IRS placed AI-powered tax scams on its 2026 Dirty Dozen list and told taxpayers plainly not to rely on AI-generated answers to complex tax questions.1Internal Revenue Service. Be Aware of Dirty Dozen Tax Scams for 2026 No software disclaimer changes who is on the hook.
What You Owe When AI Gets Your Return Wrong
Your signature on a return means you vouch for its accuracy. If an AI tool invents a credit, miscalculates your adjusted gross income, or misapplies a deduction, the deficiency notice comes to you. You owe the unpaid tax plus interest running from the original due date, regardless of which algorithm produced the number.
On top of the tax, a 20% accuracy-related penalty applies to underpayments caused by negligence or a substantial understatement of income tax. The IRS defines negligence as failing to make a reasonable attempt to follow the tax laws when preparing your return.2Internal Revenue Service. Accuracy-Related Penalty Copying whatever a chatbot produces without checking the numbers fits comfortably inside that definition. The penalty applies to the portion of the underpayment attributable to the error, not the full tax bill.3Office of the Law Revision Counsel. 26 U.S. Code 6662
You can escape that penalty by showing reasonable cause and good faith, evaluated case by case, with the most important factor being your effort to determine the correct tax.4eCFR. 26 CFR 1.6664-4 Buying and running software is not the same thing as making that effort. In Dealers Auto Auction of Southwest LLC v. Commissioner (2025), the Tax Court rejected a reasonable-cause defense where the taxpayer used software marketed as “approved for use by the IRS” but never verified the output or monitored whether filings were correct. Relying on software without internal checks, the court found, does not show ordinary business care and prudence. For an AI-assisted return, that means reviewing the output before filing and understanding your own situation well enough to spot obvious errors.
If the return crosses from negligent into fraudulent, the civil fraud penalty rises to 75% of the underpayment attributable to fraud.5Office of the Law Revision Counsel. 26 U.S. Code 6663 Criminal exposure depends on the conduct. Filing a return you know contains false statements is a felony carrying fines up to $100,000 and up to three years in prison.6Office of the Law Revision Counsel. 26 USC 7206 Delivering a document you know to be fraudulent is a separate misdemeanor with fines up to $10,000 and up to a year in prison.7Office of the Law Revision Counsel. 26 U.S. Code 7207 Willful blindness — deliberately never reading what the AI produced before signing — can satisfy the knowledge element these charges require.
When AI Companies Become Tax Return Preparers
The Internal Revenue Code defines a tax return preparer as anyone who prepares a return for compensation, and preparing a “substantial portion” counts the same as preparing the whole thing.8U.S. Government Publishing Office. 26 USC 7701 The statute exempts anyone providing only “typing, reproducing, or other mechanical assistance.” Traditional tax software has long fit inside that exemption by acting as a structured questionnaire that drops user inputs into the right form lines.
AI complicates that. A system that reads your documents, identifies deductions you didn’t ask about, or recommends how to characterize income is doing more than mechanical assistance. If a court or the IRS concludes an AI product has crossed the line, the company becomes subject to preparer penalties. For an understatement caused by an unreasonable position the company knew or should have known about, the penalty is the greater of $1,000 or 50% of the fee earned on that return. For willful or reckless conduct, it rises to the greater of $5,000 or 75% of the fee.9Office of the Law Revision Counsel. 26 USC 6694 Per-return figures multiply quickly across a customer base measured in millions.
The courts have not definitively resolved whether current AI tax tools are preparers or calculators. The more autonomous the product, the harder the mechanical-assistance argument gets. Companies marketing their tools as replacements for a human tax professional are, in effect, arguing themselves into preparer status.
Circular 230 Duties for Practitioners Using AI
Attorneys, CPAs, and enrolled agents who use AI in their practices face a separate layer of rules under Treasury Department Circular 230. Circular 230 governs anyone who represents clients before the IRS and imposes due diligence obligations that do not disappear because software handled the math.10Internal Revenue Service. Treasury Department Circular 230
The due diligence standard requires practitioners to take reasonable steps to ensure the accuracy of returns they prepare and any representations they make to the IRS or to clients. Running a client’s documents through an AI tool and filing whatever comes back, without independent verification, almost certainly falls short. The regulation looks at what the practitioner actually did to confirm correctness, not what the software claimed it would do.
There is an open debate about whether companies building tax-focused AI could themselves fall under Circular 230, particularly when their tools give specific advice rather than compute a figure. If an AI system recommends a position lacking a realistic possibility of being sustained on its merits, the company could face administrative sanctions. Penalties under Circular 230 include public censure, suspension, permanent disbarment from IRS practice, and monetary penalties capped at the gross income derived from the offending conduct.11Internal Revenue Service. Office of Professional Responsibility and Circular 230
Unauthorized Practice of Law
Every state restricts who can practice law, and tax advice often shades into legal territory. Basic tax software stays clear by functioning as a questionnaire. The risk climbs when AI starts interpreting statutes, advising on how to structure transactions, or recommending strategies to reduce liability. That work looks a lot like practicing law, and doing it without a license invites enforcement.
State regulators investigate companies whose services resemble personalized legal counsel. Consequences typically include cease-and-desist orders and civil actions from consumer protection agencies or state bar associations, and penalties often include disgorgement of fees collected during the period of unauthorized practice. Some states treat it as a misdemeanor with fines or short jail sentences. The FTC’s 2024 action against DoNotPay, which marketed itself as “the world’s first robot lawyer,” shows how this plays out federally. The company settled for $193,000 and agreed to stop claiming its AI could substitute for any professional service without evidence to back that up.12Federal Trade Commission. FTC Announces Crackdown on Deceptive AI Claims and Schemes
The line between information and advice matters here. Telling a user what the standard deduction is for their filing status is information. Telling them whether to itemize based on their specific financial picture is advice. Products that blur that line draw scrutiny from both state bar authorities and federal regulators.
Taxpayer Data Rules
Tax returns hold some of the most sensitive information a person generates: Social Security numbers, income details, bank accounts, employment history. Two separate frameworks govern how it must be handled, and AI companies that process tax data answer to both.
Federal Tax Return Confidentiality
Federal law makes it a crime for anyone in the business of preparing returns to knowingly or recklessly disclose information a taxpayer provided for preparation, or to use that information for any purpose other than preparing the return. Violations are a misdemeanor punishable by up to $1,000 in fines and up to one year in prison.13Office of the Law Revision Counsel. 26 USC 7216 Civil penalties run $250 per unauthorized disclosure or use, capped at $10,000 per calendar year, and rise to $1,000 per instance with a $50,000 annual cap if the misuse is connected to identity theft.14Internal Revenue Service. Tax Preparer Penalties
This is where AI raises a genuinely new question. Training a model on taxpayer data, even to improve the product, may count as using the information for a purpose beyond return preparation. If financial histories feed a model that also serves other customers or products, the company may be violating the rule at scale, with each taxpayer’s data treated as a separate violation.
Gramm-Leach-Bliley and the FTC Safeguards Rule
The Gramm-Leach-Bliley Act requires financial institutions to explain their data-sharing practices and safeguard sensitive customer information.15Federal Trade Commission. Gramm-Leach-Bliley Act The FTC’s Safeguards Rule classifies tax preparation firms as financial institutions, so AI-powered tax services must maintain a written information security program with administrative, technical, and physical safeguards fit to the sensitivity of the data.
Falling short triggers FTC enforcement. As of 2025, the maximum civil penalty is $53,088 per violation, adjusted for inflation each January.16Federal Trade Commission. FTC 2025 Civil Penalty Amounts In a breach affecting many users, each affected record can be treated as a separate violation. State privacy laws add another layer, often letting consumers opt out of data collection and request deletion.
Deceptive AI Marketing
The FTC has said marketing AI capabilities you cannot deliver is an enforcement priority. In September 2024, it launched Operation AI Comply, a sweep against companies using AI hype for deceptive or unfair practices. The DoNotPay complaint alleged the company never tested whether its AI output matched a human lawyer’s work and never employed or retained any attorneys. The consent order bars it from claiming the AI can replace any professional service without supporting evidence.
For AI tax companies, the message is direct. Marketing a product as equivalent to a human tax professional requires testing data to back the claim. Slogans like “AI-powered accuracy” or “better than a CPA,” without real-world performance testing, are the kind of representations the FTC has already shown it will pursue, and violations of an FTC order can run up to $53,088 each.
What the IRS Has Said About Relying on AI
The IRS’s 2026 Dirty Dozen list flags AI-enabled threats including AI-powered impersonation scams using voice cloning and deepfakes, along with misleading tax advice circulating on social media. The agency has said plainly that taxpayers should not rely on AI-generated responses to complex tax questions and should verify any calculations or information an AI provides.
The Dirty Dozen list also flags “too good to be true” tax advice, warning that acting on inflated promises of deductions or credits can lead to refund delays, audits, and penalties. AI chatbots are prone to this kind of output because they are built to sound confident and helpful even when the correct answer is that you do not qualify. Because the reasonable cause defense turns on whether you exercised ordinary care, ignoring the IRS’s own published warnings about AI accuracy weakens any later argument that your reliance was reasonable.17Office of the Law Revision Counsel. 26 U.S. Code 6664 Use the tools if they help, but read what they produce before you sign it.