IRS Criminal Investigation: Tax Crime Investigation Defense

IRS Criminal Investigation: Tax Crime Investigation Defense

The Conviction Rate and What It Permits

An 89 percent conviction rate should end most conversations about whether to cooperate with IRS Criminal Investigation. It does not. The number, drawn from IRS-CI’s own fiscal year 2025 annual report, reflects the agency’s selectivity more than its strength: CI initiates cases it expects to win, declines the rest, and prosecutes with a patience that most defendants do not recognize until the trial is already decided. The result is an enforcement posture that rewards silence, punishes improvisation, and treats the initial interview as the most consequential moment in the entire case.

What the rate conceals is more useful than what it announces. Cases are declined, referred to the civil side, or resolved without indictment in numbers the annual report does not emphasize. For the person contacted by a special agent, the question is not whether CI wins at trial but whether the case reaches trial at all. That distinction governs everything a competent defense attorney does in the first ninety days.

CI referred 2,043 cases for prosecution in fiscal year 2025. Of those, 1,611 ended in conviction. The arithmetic is not complicated, though the space between referral and conviction is where defense work occurs. A prosecution recommendation is not an indictment. An indictment is not a conviction. Each stage has its own vulnerabilities, its own decision-makers, and its own points of intervention that vanish if the defense arrives late.

The Investigation You Do Not See

Before a special agent presents credentials, the case has been developing for months and in some instances for years. CI investigations are not reactive. They begin with a primary investigation, an internal review of information gathered from civil auditors, Bank Secrecy Act filings, law enforcement referrals, or the public. If the primary investigation yields sufficient indication of criminal conduct, a supervisory chain approves what the Internal Revenue Manual terms a “subject criminal investigation.” The taxpayer is not notified. There is no letter. There is no phone call.

The IRS has no legal obligation to inform a taxpayer that an audit has become a criminal matter. A revenue agent examining deductions in a civil audit may encounter what the Manual describes as firm indicators of fraud: omissions of entire income sources, fictitious deductions, two sets of records, substantial understatements with no credible explanation. Under IRM 25.1.2, the agent consults a fraud referral specialist and, if the indicators hold, the case moves to CI. The civil audit may continue during this transition, and the taxpayer’s representative, if not attuned to the signals, may continue producing documents and answering questions as though the stakes have not changed.

The signals are specific. A revenue agent who has been responsive stops returning calls. A scheduled appointment is cancelled without explanation. A second agent appears at a meeting, or the agent begins asking questions about intent rather than documentation. In the most dangerous scenario, the civil audit proceeds while a parallel criminal investigation operates beneath it: the reverse eggshell audit, in which CI is already constructing a case and the civil examination serves as a continuing source of information the taxpayer believes is being disclosed in a non-criminal context.

We treat the abrupt cessation of communication from a revenue agent as a signal that warrants immediate reassessment. Not every silence means a referral. But the cost of assuming it does not, and being wrong, is measured in years. When communication from the civil side goes quiet, we advise our clients to stop producing documents until we can determine whether the case has crossed the line.

There is a particular quality to the silence after a civil agent disappears. It is the quiet of a case being transferred, and by the time the special agent calls, the government has already organized what the taxpayer spent months unknowingly providing.

The first contact from a CI special agent typically arrives in person. Two agents. One to conduct the interview, one to observe and record. They will identify themselves, produce credentials, and read what amounts to a modified Miranda warning: a non-custodial advisement that the taxpayer has the right to remain silent, the right to an attorney, and that anything said may be used in a criminal proceeding. This is not a formality. It is the moment the government hopes to collect statements that will become the spine of its case.

Speak to no one. Not to the agents. Not to a business partner. Not to the accountant who prepared the return. The only conversation that should follow the appearance of a CI special agent is with counsel who practices in this area, conducted under the protection of attorney-client privilege. The accountant-client privilege under IRC Section 7525 does not extend to criminal matters.

Willfulness

Every federal tax crime of consequence requires proof of willfulness: the voluntary, intentional violation of a known legal duty. The Supreme Court addressed the standard in Cheek v. United States, 498 U.S. 192 (1991), a case involving a commercial airline pilot who stopped filing returns after attending anti-tax seminars. The Court held that a good-faith misunderstanding of the tax law, even one that is not objectively reasonable, negates the willfulness element. A person who genuinely believed, however incorrectly, that wages were not taxable income under the Code could not be convicted of willfully evading tax on those wages.

Cheek is the defense attorney’s anchor, though it is not as generous as it first appears. The defense requires the defendant to have held the belief sincerely, and juries are permitted to consider the reasonableness of the belief as evidence of whether it was sincerely held. A defendant who structured transactions to conceal income, maintained dual records, or deposited funds in nominee accounts will struggle to persuade twelve people that the failure to report was an innocent misunderstanding. The pattern of conduct becomes the government’s best evidence of intent.

In practice, the willfulness defense has force when the complexity of the Code itself generated the error. An aggressive but plausible interpretation of a deduction provision, reliance on professional advice that turned out to be wrong, a genuine misapplication of the rules governing foreign account reporting. We regard the distinction between following bad advice and ignoring good advice as the line that separates cases CI declines from cases it prosecutes.

Whether the willfulness standard can bear the weight courts have placed on it is a question I cannot resolve here, though the answer matters less than the practical reality: juries convict when the conduct looks deliberate, regardless of what instruction the judge reads.

The Eggshell Problem

The eggshell audit is the scenario that most tax practitioners will encounter before their career concludes. The return contains errors the client knows about. The civil auditor does not. The question is not whether to correct the record but how to do so without triggering a criminal referral.

This is where representation matters more than doctrine. The original return preparer should never represent a client in an eggshell audit. The preparer lacks attorney-client privilege, is potentially a witness for the government, and has a personal interest in minimizing their own exposure that may conflict with the client’s defense. The Internal Revenue Code’s accountant-client privilege under Section 7525, limited as it is, does not apply in criminal proceedings at all.

The correct structure is a criminal tax attorney directing the examination, with a CPA retained under a Kovel arrangement (United States v. Kovel, 296 F.2d 918 (2d Cir. 1961)). The Kovel agreement extends the attorney-client privilege to the accountant’s communications with the client, so long as the accountant is functioning as an agent of the attorney rather than as an independent advisor. The arrangement must be established before sensitive information is shared. Once a communication occurs outside the privilege, it cannot be retroactively protected, and the accountant can be compelled to testify.

The objectives in an eggshell audit are ordered: keep the matter civil, avoid the 75 percent fraud penalty under IRC Section 6663, minimize the tax adjustment, limit the years under examination. These objectives sometimes work against one another. A full voluntary disclosure may eliminate criminal risk but widen the scope of the civil liability. A narrow response may limit the adjustment but produce the kind of unanswered questions that generate suspicion.

We approach this calibration differently than many firms. The standard advice is to cooperate fully and correct errors on the theory that candor reduces referral risk. That advice is sound in the abstract. In our experience, it underestimates the frequency with which examiners interpret voluntary corrections as admissions of prior knowledge. An error the client corrects is also an error the client knew about. Whether the examiner treats that knowledge as negligence or as fraud depends on factors (the agent’s training and disposition, the size of the discrepancy, the examiner’s current caseload, whether the field office has recently emphasized fraud-referral metrics) that no legal framework can predict with certainty. We prefer to assess the full exposure before producing any document, identify which corrections can be framed as computational or interpretive rather than substantive, and control the pace of disclosure so that the examiner receives information in a context we have constructed rather than one the examiner assembles independently.

And there is a moment in every eggshell audit where the attorney must decide whether the risk of continued cooperation exceeds the risk of asserting the Fifth Amendment. Invoking the privilege in a civil audit does not constitute an admission of wrongdoing, though it will almost certainly trigger an adverse inference and may accelerate a referral. The timing of that decision is everything. Too early, and the client loses the opportunity to resolve the matter civilly. Too late, and the client has already provided the evidence CI will use.

The client who calls after the special agent has already visited has a case. The client who calls while the civil audit is still open has options. The difference between those two situations is usually six months and one unreturned phone call.

Six months before the FY2025 annual report was published, we had a client whose revenue agent cancelled two consecutive appointments without rescheduling. The client’s prior counsel interpreted the silence as administrative backlog. We interpreted it as a referral in progress. We stopped producing documents, retained a forensic accountant under Kovel, and began preparing for a criminal defense. CI never brought charges. I am less certain than I would like to be about whether the outcome would have changed had the client continued cooperating for another quarter; the evidence the prior counsel had already provided was not as damaging as it might have been, though it was closer to the line than anyone involved should have tolerated.

What the Conviction Rate Conceals

The 89 percent rate measures outcomes at trial and through plea agreements. It does not capture the cases CI investigated and declined to prosecute, the cases the Department of Justice Tax Division returned to CI for further development, or the cases resolved through voluntary disclosure before a prosecution recommendation was made. These invisible outcomes represent the actual terrain of defense opportunity.

IRS-CI initiated 1,380 tax crime investigations in fiscal year 2025. Of those, 834 prosecution recommendations were made. The gap between investigations opened and prosecutions recommended is the space in which a prepared defense operates. A defense attorney who engages early can shape the administrative record, present mitigating facts to the special agent’s supervisory chain, and in some cases persuade the reviewing attorney at Chief Counsel’s office that the case does not warrant referral.

Bank Secrecy Act Data and the Evidence You Cannot See

In fiscal year 2025, 94 percent of IRS-CI cases were searched against Bank Secrecy Act data, producing nearly four million queries against suspicious activity reports and currency transaction reports. The BSA filing system is the single most important evidentiary source in modern tax investigations, and it operates almost entirely outside the taxpayer’s awareness.

Financial institutions file a currency transaction report for every cash transaction exceeding ten thousand dollars, and an aggregated report when multiple transactions in a single business day cross the same threshold. In CI cases opened during fiscal year 2025, the median transaction amount on currency transaction reports fell between twelve and thirteen thousand dollars. The structuring of cash transactions to avoid reporting thresholds remains one of the most common bases for criminal charges. The threshold has not changed since 1970. The technology employed to identify structuring patterns has changed in ways that the threshold figure alone does not communicate.

Suspicious activity reports are filed at the institution’s discretion when a transaction appears unusual, inconsistent with the customer’s known activity, or suggestive of potential criminal conduct. The taxpayer is never notified that a SAR has been filed; federal law prohibits the institution from disclosing the report’s existence. CI may be reviewing detailed descriptions of a taxpayer’s financial conduct (the bank’s narrative of why the activity appeared suspicious, the transaction amounts, the counterparties, the dates) long before the taxpayer has any indication that an investigation exists.

CI’s new feedback initiative, CI-FIRST, has accelerated this process. By providing financial institutions with structured guidance on what types of activity are most useful to investigators, CI is training the banking system to produce more precise reporting. The quality of SAR narratives, which in prior years were often boilerplate, is improving in direct response to this guidance. For the taxpayer who believes that a cash business or an offshore account exists beneath the government’s awareness, the assumption is almost certainly incorrect. The question is not whether the government possesses the data but whether it has decided to act on it.

An IRS criminal investigation begins long before the taxpayer becomes aware of it. By the time a special agent appears, the quiet has been productive for months.

The taxpayer who contacts counsel before that quiet resolves into a visit occupies a different position than the one who waits. Early engagement permits the construction of a defense before the government’s theory has hardened, before the administrative record is complete, before statements made to civil auditors have been recharacterized as evidence of intent. The distinction between a civil resolution and a criminal prosecution is often determined not by the severity of the underlying conduct but by the timing of the response.

A consultation is the beginning of that response. It costs nothing, assumes nothing, and establishes the privilege that protects everything discussed within it. The IRS does not abandon investigations. It completes them.