United States / government-wide
Government-wide federal fraud estimate
GAO estimated annual federal fraud losses across government. Its model covers detected and undetected fraud, with substantial uncertainty.
Statistical fraud estimate; agency disagreement
Ask Impact Analyst
Ask about the numbers or evidence. Keep the conversation going with follow-up questions.
By sending, you agree to share your question and this case’s notes with Google Gemini and save the conversation privately for review. Privacy
What was the money for?
Protect federal funds from fraud while delivering authorized services.
GAO annual statistical fraud-loss estimate using FY2018–FY2022 data. It is not a five-year sum or a forecast.
Exact amount, time period and source wording
GAO annual statistical fraud-loss estimate using FY2018–FY2022 data. It is not a five-year sum or a forecast.
Period: FY2018–FY2022 analysis; published 2024-04-16
Original plan: Protect federal funds from fraud while delivering authorized services.
Putting the amount in context
of average federal obligations in GAO’s model
The modeled annual loss range is equivalent to roughly $3–$7 per $100 of average obligations in the studied period.
What this comparison means: This government-wide relationship cannot be applied to any individual agency or program, and it does not forecast future losses.
How we calculated this and where it comes from
This government-wide relationship cannot be applied to any individual agency or program, and it does not forecast future losses.
What actually happened?
Government-wide model with uncertainty. The model is not a transaction ledger. Individual payments, detection dates and preventable losses remain unidentified.
Keep in mind: OMB disagreed with the estimate. Do not apply the range to individual agencies, project future losses or add overlapping case estimates.
Read the detailed findings
GAO estimated annual federal fraud losses across government. Its model covers detected and undetected fraud, with substantial uncertainty.
Reported status: Government-wide model with uncertainty
OMB disagreed with the estimate. Do not apply the range to individual agencies, project future losses or add overlapping case estimates.
Can we follow the payments?
A report can tell us the total. To follow the money, we also need to connect each bill to a payment and to what was received.
Government-wide scale · historical illustration
What does a trillion-dollar scenario require?
GAO estimates $233–521 billion in annual federal fraud losses using FY2018–FY2022 data. ImpactLine multiplies the annual range by five to illustrate the scale over that historical window. The five-year total is derived here. GAO did not publish a monthly loss ledger, and OMB disputed its annual estimate.
Assumes action after 3 months could prevent 100% of the remaining amount. Effectiveness remains unmeasured. The five-year window is fixed to FY2018–FY2022.
Government-wide scope overlaps individual federal cases and programs. This result is never added to the case totals, project payments or improper-payment estimates. It cannot be assigned to individual agencies.
See the calculation, dates and evidence limits
2017-10 through 2022-09 · 60 months
$1.17 trillion – $2.61 trillion × 57 / 60 × 100%
Assumes the same annual loss rate across five years, equal monthly losses and effective intervention after the selected delay. Actual monthly losses remain unknown. GAO cautions against using its estimate as a forecast. OMB disagreed with the estimate.
Net savings require evidence of the costs actually avoidable, recoveries, repayment obligations and the cost of oversight. Essential services and payments legally owed must remain protected.
Publish the payment trail, independently check progress and pause future unearned releases when the evidence warrants it.
See the Receipts Act proposal ↗Complete invoice-to-payment links
This is a government-wide statistical estimate of fraud losses. It is not a total of actual payments.
0 matching invoice and payment records linked here.This measures records linked in this file, not every record government holds, and does not establish fraud.
Help make invoices, payments and proof of delivery public.
Support the Receipts ActHow this is measured and which records are linked
Receipt matching measures this collection only. The reported estimate is not a payment denominator.
Only invoice and payment records tied to this same amount and scope count. Delivery coverage also requires a linked acceptance record. A report showing a total does not count as an individual receipt.
A positive, comparable dollar total is required to calculate coverage.
- Invoice, payment and delivery linked
- $0
- Invoice and payment linked; delivery not linked
- $0
- No matched invoice and payment in this file
- No comparable total
0 matching invoice/payment records; 0 also link to delivery.
No individual invoice-to-payment records are registered in this file yet. This does not establish that those records do not exist elsewhere.
What is still missing?
A reconciled ledger of adjudicated and suspected fraudulent payments and their dates.
Which records would help answer this?
The model is not a transaction ledger. Individual payments, detection dates and preventable losses remain unidentified.
- A reconciled ledger of adjudicated and suspected fraudulent payments and their dates.
- Dates when controls detected each scheme, action taken and recoveries.
These records have not all been assembled here. They may exist elsewhere. Personal details should stay protected.
Can we estimate how much was wasted or stolen?
This file does not establish an independent value for everything delivered. Subtracting a guessed value from a headline amount would not establish fraud. A sound estimate needs matching costs, delivered results and evidence about why any money was lost.
Sources and research notes (1 documents)
These documents support the case. A source document can describe many payments; it is not itself an itemized receipt.
- GAO government-wide fraud estimate and limits
gao.gov · 2024-04-16
Research cutoff: November 9, 2026. Older reports may not include later changes. Read our methodology.