The uncertainty band is itself a major finding
GAO estimated annual federal tax-fraud losses between $116 billion and $304 billion using IRS information from 2018 through 2024, tax-gap research and estimates of deliberately hidden economic activity. What’s the Scoop With Broach calculated that the upper bound is $188 billion above—and 2.62 times—the lower bound.
Applied to tax year 2022, GAO said the range equals about 2% to 6% of estimated tax liability. That four-percentage-point span reflects the difficulty of measuring conduct designed to remain concealed; it is not a count of adjudicated fraud cases.
| Measure | Finding |
|---|---|
| Annual estimated loss | $116B–$304B |
| Range width | $188B |
| Share of 2022 tax liability | About 2%–6% |
| Agencywide antifraud strategy | Not developed |
| Designated antifraud entity | Not designated |
Existing controls do not equal coordinated governance
IRS divisions screen returns, audit taxpayers and refer suspected crimes. IRS told GAO that its Return Review Program prevented $88 billion in invalid and potentially fraudulent refunds from 2018 through 2024.
That $88 billion should not be subtracted from the annual loss range: it covers seven years, includes invalid refunds that may not all be fraud, and measures stopped payments rather than unpaid tax. GAO nevertheless found IRS lacks both a documented agencywide antifraud strategy and a lead entity to coordinate the work.
IRS only partly agreed
GAO recommended a documented strategy and a designated antifraud entity. IRS partially agreed with both recommendations; GAO kept them open and said coordinated oversight would improve the government’s response to losses on this scale.
The document supports a governance finding, not a claim that any specific taxpayer committed fraud or that every dollar in the modeled range can be recovered. Most taxpayers comply, and fraud is a narrower concept than all nonpayment in the tax gap.
Method and limitations
The newsroom reviewed GAO’s September 25 report page and methodology description, then recomputed the range width and ratio. GAO used a probabilistic simulation because direct observation of hidden fraud is not possible.
The estimate depends on assumptions, older tax-liability data and multiple source types. It is best read as a bounded planning estimate, not a precise annual invoice.