562,000 Loans, Zero Reviewed

On April 24, 2026 the SBA referred 562,000 suspected fraudulent loans worth 22.2 billion dollars to the Department of Treasury for collection. The agency called it the largest referral package on record and wanted applause. Here is the question that detonates the whole press release. How many of those 562,000 files did a human being open and judge before the referral went out? The same number the agency checked on the way in. Zero.

Published June 20, 2026 • Filed under: The Machine Decides Now

A dense wall of electronic circuitry and glowing data nodes, representing the automated SBA referral and Treasury offset machine that flagged 562,000 borrowers for 22.2 billion dollars in collection without individual review

The SBA wants this one read as a victory lap, and it is a big number, so the lap is long. On April 24, 2026 the agency announced it had referred 562,000 suspected fraudulent loans to the U.S. Department of Treasury for collection, totaling 22.2 billion dollars in delinquent Paycheck Protection Program and COVID Economic Injury Disaster Loan debt. The agency called it its largest referral package on record. The borrowers were also transmitted to the Department of Justice, and the whole thing was announced in coordination with the White House Task Force to Eliminate Fraud. Tough agency, staggering theft, accountability at last. And buried inside that triumphant figure is the part nobody at the podium will say out loud, which is that the word referral is doing an enormous amount of work.

Because a referral is not a review. A referral is a list. The SBA did not open 562,000 files, read 562,000 applications, weigh 562,000 individual cases of suspected fraud, and decide one by one who deserved to be sent to collection. It ran the database, pulled the flags, and shipped the entire batch to Treasury in a single package. The deciding, if you can call it that, was done by the same kind of automated pattern-matching that decides every mass action this agency takes. The flag came first. The judgment, if it ever comes, comes later, and it comes from a machine that garnishes, not a person who reviews.

Nobody Checked On The Way In Either

Here is the context the SBA would love for you to forget. Between 2020 and 2021 this same agency approved roughly 1.2 trillion dollars in PPP and EIDL loans. The verification process on those loans was, to be generous, theoretical. The agency's own Inspector General estimates that at least 200 billion dollars of that money was fraudulent. Two hundred billion. That is not a rounding error or an unlucky break. That is the predictable outcome of an agency that decided speed mattered and confirmation did not, and shoveled more than a trillion dollars out the door while checking almost nothing.

So sit with the symmetry. The SBA did not check the applications on the front end, which is precisely why there are 562,000 fraudulent loans to refer on the back end. The fraud is not a thing that happened to the agency. It is a thing the agency manufactured by refusing to verify. And now, having created the mess by skipping the review, the agency is cleaning it up by skipping the review again, this time pointing the same automated indifference at collection instead of approval. The machine that never checked who got the money is now feeding the machine that never checks who gets garnished.

An agency that could not verify a single application before sending out 1.2 trillion dollars now wants credit for an automated list that sends 562,000 people to a garnishment machine. The incompetence and the enforcement are the same skill, pointed in opposite directions.

What The Treasury Offset Program Actually Does

This is where referral stops being a press-release word and starts being a real-world event in someone's bank account. The Treasury Offset Program is not a courtroom. It is an automated machine that intercepts federal money owed to you and applies it against a debt the government says you owe. Tax refunds, Social Security benefits, certain other federal payments, the offset machine reaches into all of it. There is no judge in that process. There is no hearing scheduled by default. There is a flag, a referral, and then a payment that simply does not arrive because the system took it first.

Now layer that on top of 562,000 referrals built by pattern-matching rather than individual review. The actual fraudster, the one who created a fake business to pull down free money, deserves every dollar of that offset and more. Fine. But the offset machine cannot tell the difference between a fraud ring and a legitimate borrower whose loan got flagged by the same automated indicator. It does not adjudicate. It garnishes. So the false positives, the ordinary people swept into a 562,000-name batch because their file tripped a wire, find out the same way the fraudster does, when the refund stops coming or the benefit check shrinks. The burden of proving the algorithm wrong falls entirely on them, after the money is already gone.

The Referral Is Industrial. The Review Never Happened.

Notice what the agency is bragging about. It is not bragging that it carefully investigated 562,000 cases and confirmed fraud in each one. It is bragging about the size of the package. The largest referral on record. The pride is in the volume, which is the tell, because volume is the one thing an automated batch process is good at. You can refer 562,000 loans in an afternoon if all you are doing is exporting flagged rows to a list and handing it to Treasury. You cannot review 562,000 loans in an afternoon, or a month, or a year, because review requires humans opening files, and the agency does not have, and never deployed, the army that would take.

And the loans in this package were not freshly discovered. They were flagged for suspected fraud years ago and then left sitting, never sent to Treasury for collection, never referred to DOJ. The new action is not detection. It is the decision to finally flip the switch on the offset machine for a backlog that was already flagged. That is the entire achievement being celebrated, taking an old list and pressing send. The fraud was identified by a system. The collection is executed by a system. The only human contribution to the whole 22.2 billion dollar event was the choice to point the automation at the borrowers now instead of later.

The Asymmetry, As Always

Run the agency's two speeds side by side, because this is where every LOLSBA story lands. When the SBA wants to refer a borrower to a garnishment machine, it does it at industrial scale, 562,000 at a time, no individual review, no hearing, machine-flagged and shipped. When that same borrower needs the agency to actually examine their specific file and lift a wrongful flag before the offset takes their refund, there is no timeline, no assigned reviewer, and no automated process working in their favor. The referral is industrial. The exoneration is artisanal, and it is staffed by nobody. The agency built a factory for accusing people and a suggestion box for clearing them.

That asymmetry is the product, not a glitch. The SBA spent the pandemic creating an estimated 200 billion dollars in fraud through sheer refusal to verify, and the only way to look serious after a failure that enormous is to swing the collection net as wide as the approval net once was. Wide nets catch the guilty. They also catch the innocent. And the agency has decided that the legitimate borrower garnished by mistake is acceptable collateral for a number big enough to call the largest referral on record.

The Number Is The Punchline

So here is where it lands. The SBA wants 22.2 billion dollars and 562,000 referrals to read as the agency finally getting serious about the fraud it once ignored. Read it the other way. That number is only possible because no one reviewed those files individually, not on the way in when the money left and not on the way out when the collection started. The deciding was handed to a pattern-matcher on both ends, and the consequences were handed to an offset machine that does not adjudicate. The agency that could not stop 200 billion dollars in fraud because it checked nothing is now collecting 22.2 billion of it the same way it lost it, by trusting the automation and skipping the human.

The lesson will not be learned, because the lesson is inconvenient. An agency that can refer 562,000 borrowers to Treasury in its largest package on record but could not verify a single application before it sent the money is not bad at its job. It is extremely good at the job it actually does, which is generating enormous numbers on the back end to cover for a complete absence of verification on the front end. The machine that decided who got the money never checked. The machine that decides whose refund gets seized never checks either. Same machine. Same indifference. Only the direction of the harm changes, and this time it is pointed at collection.

SHARE ON X SHARE ON FACEBOOK SHARE ON LINKEDIN