On September 25 the Government Accountability Office published an estimate that, by its own account, had no comprehensive predecessor. The report names the gap it set out to fill: "there is no comprehensive estimate of the amount of U.S. federal tax revenue lost to fraud."
Its answer: the federal government loses an estimated $116 billion to $304 billion a year to tax fraud, based on data and information from 2018 through 2024. In fiscal year 2025, the report says, the IRS collected about 84 percent of all federal revenue, more than $5 trillion.
Applied to tax year 2022, the latest the IRS has projected, the range is approximately 2 to 6 percent of all tax owed and approximately 17 to 43 percent of the gross tax gap, the roughly 15 percent of tax owed that was not paid voluntarily and on time. For tax years 2018 through 2021, footnotes put those ratios at 2 to 8 percent and approximately 17 to 56 percent.
GAO also finds the IRS has no antifraud strategy and no designated office in charge of fraud risk.
The two ends of the range come from two different models
The headline reads like one confidence interval. It is not. GAO modeled fraud three ways with Monte Carlo simulation: the IRS's own data on adjudicated and detected potential fraud; the share of the tax gap attributable to fraud, guided by the UK's tax gap data on fraud-related behavior; and tax evasion in the shadow economy, legal activity purposely hidden from the government. Selected parts of the first two were combined; the third was kept apart. A note under the report's Figure 21 gives the results:
The middle 90 percent of values for the combined estimate is $116 billion (5th percentile) and $253 billion (95th percentile). The middle 90 percent of values for the shadow economy estimate is $122 billion (5th percentile) and $304 billion (95th percentile).
The $116 billion floor is the low end of one model; the $304 billion ceiling is the high end of another. The figure says the range was assembled exactly that way, that the shadow-economy estimate ran roughly $10 billion to $50 billion above the combined one, and that about 10 percent of scenarios judged extreme were first removed from each area.
GAO calls the range its best estimate and does not claim it as a bound. The simulation "should be interpreted carefully, with an appreciation for the uncertainty associated with this methodology, which cannot be quantified." And: "we cannot eliminate the possibility that the actual amount of fraud could be outside the range of our estimate."
It includes potential fraud, in GAO's own vocabulary
In the report, fraud in the strict sense is determined through a court or adjudicative system. Everything else is a separate category: "We refer to instances of fraud that are not adjudicated as potential fraud." The estimate covers adjudicated fraud, detected potential fraud and undetected potential fraud. The top of the range comes entirely from the shadow-economy model, which the report labels undetected potential fraud.
One stated assumption is that "IRS data on detected potential fraud may include some instances of nonfraudulent activities." The model sets the loss to zero for a portion of those cases, and the shadow-economy model carries a parameter for how much hidden activity is intentional evasion. That boundary is the subject of the IRS's first objection.
What is in the number and what is not
- Only IRS-collected tax. Fraud in customs duties, including tariff revenue collected by the Department of Homeland Security, is not counted; neither is fraud in fees and other nontax revenue collected by more than 30 agencies.
- Gross, not net. "This estimate does not account for fraud losses that IRS may have ultimately recovered, or may recover in the future, through collection, litigation, or restitution."
- Identity theft and detected criminal fraud are in, though neither is in the IRS's own tax gap estimates.
- Not a forecast. GAO: "our estimate cannot be used to predict the exact amount of future federal tax fraud loss."
- Older foundations. Many components of the IRS tax gap projections for 2018 through 2022 rest on the IRS's tax year 2014-2016 estimates.
What the IRS already stops
By IRS figures, its Return Review Program, which screens individual returns claiming refunds, prevented approximately $88 billion in invalid and potentially fraudulent refunds from 2018 through 2024. That is a seven-year total; the loss estimate is per year. Auditors over the same years recommended an annual average of $24.9 billion in additional assessments, figures GAO notes cover all noncompliance, not only fraud. In fiscal year 2024 the IRS reported assessing 1,400 civil fraud penalties totaling $115 million and initiated more than 2,600 criminal investigations.
No strategy, and no one designated to own one
GAO describes the IRS's approach to fraud risk management as "ad hoc and managed by individual IRS divisions." The IRS has assessed its fraud risks in line with leading practice, GAO says, but has no strategy for how its controls answer them. Its risk office told GAO there is no antifraud strategy and "ownership of all antifraud programs and mitigations remains with each division." The divisions have none either.
Asked why, some division officials believed the risk office ran an agency-wide strategy; that office said none exists. Others said they "had not identified a specific need for an antifraud strategy", or that "their division's mission statement represented their antifraud strategy."
GAO's example of the gap: in 2023 the IRS identified a new fraud risk, "the risk that the enactment of new tax credits before the implementation of effective fraud controls would reduce tax revenue and erode public confidence in the integrity of the tax system." It "did not identify a specific plan to address it."
The recommendations and the reply
GAO made two recommendations:
The Commissioner of Internal Revenue should develop and document an agency-wide antifraud strategy, or direct divisions to develop and document antifraud strategies at the operating level with oversight by the designated antifraud entity.
The Commissioner of Internal Revenue should designate an antifraud entity responsible for coordinating and overseeing fraud risk management activities.
The IRS partially agreed with both, in a letter dated September 10 and signed by Frank J. Bisignano as chief executive officer. As GAO renders it, the IRS said it would keep documenting its efforts through existing programs and "consider developing an agency-wide antifraud strategy". It named its Chief Tax Compliance Officer as the entity coordinating division fraud work. GAO called that "a positive step toward addressing our recommendation" but said that to fully implement it, the IRS should ensure that office takes on, and documents, all of an antifraud entity's responsibilities.
The IRS objected that the report does not sufficiently distinguish fraud from broader noncompliance. GAO: "However, this is not the case. Our estimate of tax fraud loss excludes losses due to instances of noncompliance that are not fraud." The IRS also said that calling its fraud governance fragmented does not reflect its approach, and that more analysis would help show how centralized oversight would materially improve outcomes. GAO pointed to the IRS's own $606 billion net tax gap projection for tax year 2022, and said preventive controls are "more cost-effective than a pay-and-chase approach that relies on detecting fraudulent activity and attempting to recover funds."
What comes next
No deadline is attached to either recommendation. On September 27 GAO's product page listed both as open.
An older recommendation is still waiting. In 2024 GAO asked Treasury to find ways to expand fraud estimation across government; "As of July 2026, we are awaiting an update from Treasury about the status of its actions in response to this recommendation." That 2024 work, GAO-24-105833, put government-wide fraud losses at $233 billion to $521 billion a year. This report cites it but does not say whether the two ranges overlap or can be added.
What the report does not say
- How much of the range is adjudicated fraud and how much is potential fraud.
- Which taxpayers, industries, schemes or tax types account for how much of the loss.
- Who is right about the line between fraud and noncompliance; the dispute is unresolved.
- Any date by which the IRS must act.
- Anything about any individual taxpayer.