9
There’s a federal statute that can turn a tax-preparer’s mistake into a criminal charge — and most CPAs have never heard of it. IRC §7216 makes it a misdemeanor to knowingly or recklessly disclose a taxpayer’s return information without consent, and a felony for a second offense. The penalties are real: fines and up to a year in prison per violation.
What §7216 actually covers
The statute is broader than most practitioners assume. It reaches any disclosure of “tax return information” — not just the return itself, but the underlying data a preparer collects to complete it: income figures, deductions, dependents, account details, and the notes a firm keeps on a client’s situation. The trigger is disclosure to a third party without the taxpayer’s informed consent. And the standard is not intent to harm. It is knowingly or recklessly making the disclosure at all.
That is the part that should make every CPA firm sit up. The statute does not require malice. It requires a disclosure that a reasonable preparer should have known was unauthorized. In the ordinary course of tax season, that line is easy to cross without ever meaning to.
What public AI does to the calculus
Now consider what happens when a preparer pastes client return data into a public AI tool to draft a response or check a calculation. That data is transmitted to a third party’s infrastructure — a disclosure of return information. Whether it’s “knowing” is a question no firm wants to be the test case for.
The tool may retain the prompt for training. It may route the data through servers in jurisdictions with different rules. It may be accessible to the provider’s staff. Each of these is a third party receiving return information without the taxpayer’s consent. The firm did not intend to violate the statute. But the disclosure happened, and the record of it now lives outside the firm’s control.
Why the fix is architectural
The fix isn’t to stop using AI in tax season. That would be throwing away a real productivity gain. The fix is to use AI that never transmits the data anywhere — open-weights models running on infrastructure you control, where the return information stays inside your four walls.
When the model runs on-premises, nothing is disclosed to a third party. There is no transmission, no retention on a vendor’s servers, no chain of custody you cannot inspect. The taxpayer’s return information stays where it belongs, under the firm’s own controls, subject to the firm’s own audit trail. The capability is the same. The exposure is not.
The question every firm should ask
Before tax season, every CPA firm should ask a single question: does any client return data leave our walls when we use AI? If the answer is yes — or if you cannot answer with certainty — you are carrying §7216 exposure you may not have priced in.
If you’re a CPA firm using public AI on client data, this is the statute worth reading before tax season. Start with the Confidential AI Exposure Checklist to see where you stand. Nine lives. Zero leaks.