How Treasury Helped Catch $11.7 Billion in Bad Payments, and the Three-Part Check a Small Business Can Copy

People search for "autonomous fraud detection" as if it were one machine that watches the money and decides on its own. The best public example in the United States says otherwise. On September 23, 2026 the Treasury's Bureau of the Fiscal Service reported what its payment screening did in fiscal year 2025, and the numbers show a system built from simple checks, good data and people who act on what the checks find.

What Treasury reported

In fiscal year 2025, Treasury helped prevent, detect, and recover an estimated $11.7 billion in fraud and improper payments, a 38% increase from the previous year (Bureau of the Fiscal Service, September 23, 2026). The work sits in the Office of Financial Integrity, which turned the Do Not Pay system into a central fraud prevention resource for federal agencies and states.

The same release gives the scale. Do Not Pay screenings rose 216% to 641 million. State partnerships found more than $1.5 billion in duplicate or improper payments. Testing 19 new datasets revealed $28 billion in potential improper payments that can be stopped in the future.

What "autonomous" looks like inside

An earlier Treasury release breaks down how the machinery works (U.S. Treasury, press release jy2650, October 2024). In fiscal year 2024, expanding risk-based screening prevented $500 million. Identifying and prioritizing high-risk transactions prevented $2.5 billion. Using machine learning to identify Treasury check fraud faster recovered $1 billion.

Read those three lines together and the pattern is clear. First, every payment is screened before it leaves. Second, the riskiest payments are ranked first, so human attention goes where the loss would be largest. Third, machine learning is used where the volume is too high for people, in this case finding fraudulent checks. In this design the model finds, and people decide.

Treasury's own fraud executive put it plainly in the 2026 release: preventing fraud requires more than technology. It requires strong data, strong partnerships, and stopping problems before the money goes out the door.

Why this matters for a small business

The other side of the ledger is getting worse. According to the Federal Trade Commission, people reported losing about $16 billion to fraud in 2025, the highest on record and about 25% more than in 2024. Nearly $1 billion of it went to scammers posing as businesses, with the largest losses to people posing as banks (FTC, June 2026).

A small business will not build a Do Not Pay system. It does not need to. The rule behind Treasury's numbers fits on a sticky note, and the Deputy Secretary said it out loud in 2024: pay the right person, in the right amount, at the right time.

The three-part check before any payment

1. Right person. Is this vendor already on file, with the same name and the same bank account as last time?
2. Right amount. Does the amount match the invoice or contract, and is this invoice not a duplicate of one you already paid?
3. Right time. Is the payment due now, or is someone pushing you to pay early, today, before anyone can look?

If the bank details changed, do not use the phone number or email in the request. Call back on a number you already know, the same advice the FTC gives for voice scams: use a phone number you know is theirs.

These checks are exactly what software can run automatically on every payment, and flag the exceptions for a person to approve. That is the honest version of autonomous fraud detection: the machine never gets tired of checking, and a human still signs off on the exception.

For more on how scammers now use AI voices to push urgent payments, read our breakdown of the FBI warning on voice clone fraud.

Sources. Bureau of the Fiscal Service: Treasury Teams Honored for Safeguarding Public Funds and Expanding Data Transparency, September 23, 2026. U.S. Department of the Treasury: press release jy2650, October 2024. Federal Trade Commission: FTC Data Show People Reported Losing $3.5 Billion to Imposter Scams in 2025; Scammers use AI to enhance their family emergency schemes.

PATech Labs builds AI automation for small businesses in the United States, including payment checks that run on every transaction and send the exceptions to a person for approval. If you want the right person, right amount, right time rule built into your workflow, talk to us.