AI tax advice risks for business owners are not theoretical anymore.
Back in April, I warned you about people using AI to do their taxes.
Well, here’s a cautionary tale for you.
In Clinco v. Commissioner, T.C. Memo. 2026-16, a lawyer was in Tax Court defending a client against a nearly $2.3 million IRS notice of deficiency.
He had an idea: if the IRS notice was missing a “wet” manual signature from an IRS employee, maybe the whole notice could be thrown out.
If he won, the client walked.
So the argument went into the brief.
The court filing does not name the AI tool, and the Tax Court stopped short of saying the record proved generative AI was used. But the opinion was blunt: the case names, reporter citations, and legal propositions looked like something “cooked up by AI.”
The lawyer cited four cases to support the argument.
Three appeared to be hallucinations.
One case did not exist. Another did not exist as cited. Another cited page pointed to a completely unrelated Tax Court case about S corporation management fees.
Then it gets worse.
The actual law had been settled since before World War II.
A notice of deficiency does not need a wet signature to be valid. The Tax Court pointed to case law saying even an unsigned notice of deficiency can be valid.
So the AI did not just invent the cases.
It invented the rule those cases were supposed to prove.
Is this the first time a lawyer used AI in court with disastrous results?
No.
This started showing up in 2023, almost as soon as ChatGPT was hot off the griddle.
But this was one of the first times the Tax Court had to deal directly with what looked like fake AI citations in a tax case.
The judge did not impose sanctions this time. The court basically said the embarrassment, the failure to cite-check, and the failure to explain the citations were enough for now.
But the warning was clear.
Submitting fake case law is unacceptable. Other courts have already sanctioned lawyers for this. The Tax Court has not done it yet.
Yet.
And that word matters.
Here is the part business owners should pay attention to.
That lawyer reads case law for a living.
He still filed citations that collapsed the minute the court checked them.
Why?
Because AI sounds confident.
That is the trap.
You ask a question. It gives you a clean answer. It gives you cases. It gives you a structure. It may even sound like a smart associate who has been working late and drinking too much coffee.
But unless you know how to verify the answer, nothing in the response tells you whether it is real.
That is exactly why AI tax prep risks and AI tax advice risks are different from normal software errors.
A software error looks like a broken calculation.
An AI error can look like a well-written explanation.
This is not just one embarrassing Tax Court moment.
Researchers in Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools looked at how legal AI tools performed on legal research questions.
The numbers are not comforting.
Prior research found general-purpose AI tools hallucinated on legal queries 58% to 82% of the time.
And the specialized legal AI tools — the ones actually built for legal research — still hallucinated 17% to 33% of the time.
That is the legal world.
Now layer tax on top of that.
Tax is not one rule. It is usually a stack of rules.
A real answer may need the Internal Revenue Code, Treasury Regulations, IRS guidance, revenue rulings, court cases, state law, entity structure, basis rules, timing rules, and your actual facts all working together.
AI tools can mix those layers up.
They can cite an old rule.
They can miss the exception.
They can confuse what applies to an individual with what applies to an S corporation, partnership, trust, or C corporation.
They can give you a great answer to a question you did not actually ask.
And they can do all of that with perfect grammar.
This is the deeper issue.
AI is designed to produce an answer.
Tax strategy is often about slowing down long enough to ask better questions.
For example:
That is where AI gets dangerous.
Not because it is useless.
Because it gives answers before it understands the whole board.
Look. I am not telling you to stop using AI.
I use it. My team uses it.
There are places where AI can save time:
That is useful.
But that is not tax advice.
That is support work.
There is a big difference between using AI to organize information and using AI to decide whether you should restructure your company, claim a deduction, move assets, file an election, or take a reporting position on a return.
For a $5 million to $20 million business, “the bot said so” is not a defense the IRS accepts.
It is not a defense the Tax Court accepted from a lawyer with decades of practice.
It is definitely not going to work for you.
If it goes sideways, you signed the return.
You own the position.
You own the penalty exposure.
You own the cleanup cost.
And the bigger your business gets, the more expensive a bad assumption becomes.
A missed deduction is annoying.
A bad entity move can follow you for years.
A wrong tax position can create penalties, amended returns, legal fees, delayed exits, state issues, and lost credibility.
Here is the pattern I hope you notice.
The strategy moves that put real money in your pocket are usually the ones AI gets wrong confidently.
The easy questions are not where the damage happens.
The damage happens when the answer depends on five or six facts the AI cannot see.
That includes questions like:
These are not prompt problems.
They are judgment problems.
If your business is profitable and you are using AI to research tax strategy, use it as a starting point — not the final answer.
Before you rely on a tax position, have a human being review the full chain.
Schedule a Free Tax Strategy Session if you want to pressure-test your structure, entity setup, deductions, and strategy before something expensive gets locked into a return.
AI can help you move faster.
It can help you organize.
It can help you understand the conversation before you walk into a meeting.
But it does not replace tax judgment.
It does not verify its own citations.
It does not know your facts unless you give them to it.
And even then, it may still give you a clean, confident answer that is completely wrong.
For business owners, that is the risk.
Not that AI is useless.
That it is useful enough to make you trust it in places where you should not.
Next week, I am getting back to the Section 1202 conversation — specifically, how owners restructure into a qualifying entity without setting fire to their current tax position.
That is exactly the kind of conversation I want to have mano a mano as humans.
Because when the strategy is big enough to matter, it is too big to outsource to a bot.
AI can help explain general concepts, organize information, and prepare questions. It should not be treated as a reliable source for tax advice without professional review and source verification.
Tax questions often depend on entity type, timing, income character, state law, documentation, elections, and recent law changes. AI can miss those facts or apply the wrong rule.
Yes. In Clinco v. Commissioner, the Tax Court said three of four supporting cases appeared to be hallucinations generated by a large language model AI.
AI can be useful as a support tool. It is risky when used to make final tax decisions, claim deductions, choose entity structures, or take reporting positions without professional review.
Verify the authority, check whether the law is current, confirm the answer applies to your entity and facts, and have a qualified tax professional review the position before filing or restructuring.
Schedule a Free Tax Strategy Session if you want a human review of the tax strategy before you rely on it.
Disclaimer: This article is for general educational purposes only and is not legal, tax, or accounting advice for your specific situation.