I’m Darrin Mish. Tampa tax attorney, 32 years in, more than $100 million in IRS debt resolved. What follows isn’t theory – it’s what I’ve actually watched work.
A lawyer walked into Tax Court and cited a case that did not exist. The opposing side checked. The judge checked. The case was a complete fabrication – invented by an AI tool that sounded confident while being completely wrong.
This is not a hypothetical. It is happening right now, across the country, and the consequences are brutal.
The Problem Is Bigger Than You Think
By late 2025, there were approximately 800 documented cases of AI citation errors across more than 25 jurisdictions. And that number is growing fast.
Here is what is happening. Tax professionals, attorneys, and even self-represented taxpayers are using AI tools like ChatGPT and similar platforms to research tax law. The AI generates responses that look authoritative – complete with case names, court citations, and legal holdings. The problem is that AI “hallucinates.” It makes things up. It fabricates case names, invents holdings, and creates citations to Revenue Rulings that never existed.
And it does all of this with absolute confidence. There is no disclaimer that says “by the way, I just made this up.” The fake citations look identical to real ones. The formatting is perfect. The party names sound plausible. The holding reads like something a real court would say. But the entire thing was generated from patterns in training data, not from actual legal research.
Real Sanctions, Real Consequences
The courts are not amused. Here are some examples of what has happened to people who trusted AI without verification.
In Lothamer Tax Resolution, Inc. v. Kimmel, a Western District of Michigan case from 2025, a self-represented defendant cited 29 fake cases generated by AI. The court imposed $2,900 in sanctions – $100 per fabricated citation. The judge did not care that the defendant was not a lawyer. Filing documents with the court comes with an obligation to verify what you are saying is true, whether you went to law school or not.
In a Northern District of Georgia matter, an attorney submitted a brief where 17 of 24 cited cases either did not exist or were materially misquoted. Seventeen out of twenty-four. That is a 71% fabrication rate in a single filing. The attorney’s credibility with that court is effectively destroyed.
Tax Court has its own sanctions provision under IRC Section 6673, which allows penalties up to $25,000 for frivolous or groundless positions. Filing a brief built on cases that do not exist? That qualifies. And unlike regular court sanctions, Section 6673 penalties are imposed on the taxpayer, not the attorney – though the attorney can face separate professional discipline.
Why AI Gets Tax Law So Wrong
Tax law is one of the worst areas for AI research, and here is why. The Internal Revenue Code is massive – over 10,000 sections. Treasury Regulations add thousands more pages. The Internal Revenue Manual, Revenue Rulings, Revenue Procedures, Private Letter Rulings, Tax Court decisions, Circuit Court opinions – the body of tax law is enormous and deeply interconnected.
AI models are trained on general internet data. They can identify patterns and generate text that looks like legal writing. But they do not understand the law. They do not know the difference between a real Tax Court case and a law review article discussing a hypothetical scenario. They cannot distinguish between current law and provisions that were repealed 20 years ago.
So when you ask an AI tool about the qualified business income deduction under Section 199A, it might cite a “case” that sounds perfectly plausible – correct formatting, realistic party names, appropriate court – but the case simply does not exist. The AI assembled it from patterns in its training data, not from actual legal research. It is a convincing-looking forgery created by a machine that does not know the difference between real and fake.
The Tax Court’s Unique Vulnerability
Here is something most people do not realize. Over 75% of Tax Court cases involve pro se litigants – people representing themselves without an attorney. Many of these taxpayers are turning to AI for help because they cannot afford professional representation.
This creates a perfect storm. Self-represented taxpayers who do not have the training or resources to verify legal citations are relying on tools that regularly fabricate them. The Tax Court is seeing a surge in filings built on fictional authority, and judges are responding with increasing severity.
I want to be clear: I am not against using AI in tax practice. I use it myself – for drafting, for brainstorming, for organizing complex factual scenarios, for identifying potential issues I might want to research further. But I verify every single citation against primary sources before it goes into any document that leaves my office. Every single one. No exceptions.
What the Courts Expect From You
Whether you are an attorney, a CPA, an enrolled agent, or a taxpayer representing yourself, filing a document with any court carries an implicit certification that the facts and legal authorities cited are accurate to the best of your knowledge.
Federal Rule of Civil Procedure 11 requires that legal contentions be “warranted by existing law or by a nonfrivolous argument.” When you cite a case that does not exist, you have failed that standard. It does not matter that the AI told you it was real. It does not matter that you did not know it was fake. You signed the document. You are responsible.
Courts have drawn a clear line between two responses when AI errors are discovered. Attorneys who immediately admitted the error, apologized, and explained what happened received lighter sanctions. Attorneys who tried to deflect, minimize, or claim the AI must have been right? They got hammered. The cover-up is always worse than the crime.
How to Use AI Without Getting Burned
I am a practical person. AI tools are not going away, and they can genuinely save time in tax practice when used correctly. Here is how to use them without destroying your career or your case.
Never cite anything without pulling the source document. If an AI tells you about Tax Court case “Smith v. Commissioner, T.C. Memo 2022-45,” go find T.C. Memo 2022-45 on the Tax Court’s website or in a professional database. Read it. Verify it says what the AI claims it says. If you cannot find it, it probably does not exist.
Cross-reference with professional databases. Westlaw, LexisNexis, Bloomberg Tax, and the Tax Court’s own DAWSON system are your verification tools. If a case does not appear in these databases, it is not real. Full stop.
Be especially skeptical of Revenue Rulings and Private Letter Rulings. AI tools frequently fabricate these. A fake “Rev. Rul. 2019-45” sounds plausible but is easy to verify on the IRS website. Take the 30 seconds to check. It could save you thousands in sanctions.
Use AI for structure, not authority. AI is excellent at helping you organize arguments, identify potential issues, draft prose, and think through factual scenarios. It is terrible at providing reliable legal citations. Use it for the first job and never trust it for the second.
Document your verification process. If you use AI as part of your research workflow, keep records showing that you verified each citation against primary sources. This protects you if a citation error somehow slips through – you can show the court you had a responsible process in place, not that you blindly copied AI output into a filing.
The Ethical Obligation Is Yours, Not the AI’s
Here is the part that some professionals do not want to hear. AI does not have a license to practice law. You do. AI is not bound by professional responsibility rules. You are. AI cannot be sanctioned, disbarred, or reported to a disciplinary committee. You can.
The attorney’s duty of competence includes verifying the accuracy of legal authorities before presenting them to a court. The enrolled agent’s duty under Circular 230 includes exercising due diligence in preparing submissions to the IRS. The CPA’s professional standards require accuracy in tax return preparation and representation.
None of these obligations include a carve-out for “but the computer told me.” The tool you use to do your work does not change the standard you are held to. A surgeon who uses a robotic assistant is still responsible for the outcome of the surgery. A pilot who uses an autopilot system is still responsible for the safe operation of the aircraft. The same principle applies here.
What This Means for Taxpayers
If you are a taxpayer thinking about representing yourself before the IRS or Tax Court, please hear this. AI can help you understand general concepts. It can help you organize your thoughts. But if you use it to find legal authority for your position, you must verify those citations independently before putting them in any document you file.
Better yet, talk to a professional. The cost of a consultation with a qualified tax attorney is a fraction of what a $25,000 Section 6673 penalty will cost you. And it is infinitely cheaper than the damage to your case when the judge discovers your brief is built on imaginary law.
Get Help Now
If you are dealing with a Tax Court case, an IRS audit, or any tax dispute, having a real attorney who verifies real law is not a luxury – it is a necessity. Contact the Law Offices of Darrin T. Mish, P.A. at (813) 229-7100 for a free consultation.