Every so often the SBA does something so perfectly on-brand that satire just has to step aside and let the procurement document talk. This is one of those times. The agency signed a contract with Palantir, the data-analytics firm whose entire business is fusing every record about a human being into one searchable surface, to help run a nationwide hunt for pandemic loan fraud. And the contract, the actual line item, is labeled the SBA Fraud Prevention Pilot and Bootcamp, with a projected end date in early April, procured off the General Services Administration's standing schedule like it was a box of printer toner. The dollar figure on this exercise in total information awareness was $300,000.
Three hundred thousand dollars. Sit with that number, because it is doing two contradictory jobs at once. It is small enough to be a rounding error in a fraud problem measured in the tens of billions, which tells you this is a pilot, a trial balloon, a toe in the water. And it is the price of attaching the most aggressive private surveillance apparatus in existence to a federal benefits agency, which tells you the toe in the water belongs to a very large and very thirsty machine. The SBA did not buy a solution. It bought a demo of a surveillance capability, on the public schedule, and gave the demo a summer-camp name.
What $300,000 Actually Rents You
Understand what Palantir's tooling does and the word Bootcamp starts to feel like a joke told with a straight face. The Foundry platform consolidates what the company politely calls disparate information assets into a unified data environment, which in plain language means it takes your bank records, your tax filings, your corporate registrations, your property records, and your social media and stacks them into one profile. Bolted on top is a Case Manager that tracks every investigative step and automatically sorts alerts by risk level, so a human analyst never has to do the tedious work of deciding who looks guilty. The machine does the suspecting. The human does the clicking.
This is not fraud detection in the sense a normal person means it, where you flag the obviously fake application at the door. This is dragnet architecture. It is built to ingest an entire population of borrowers and rank them, to assemble a dossier on each one from sources that were never collected for this purpose, and to surface the ones the algorithm decides are risky. The exact same Foundry-shaped capability has been pointed at tax data for massive-scale mining elsewhere in the federal government. Now a slice of it has been pointed at the people who took emergency loans during a government-ordered shutdown, and the bill for the introductory session came to less than the price of a modest house.
The Trigger Was A Video
Here is the part that should make a taxpayer's eye twitch. The catalyst for hiring a planetary surveillance contractor was not a forensic audit, not an inspector general's multi-year investigation, not a careful statistical finding. It was a video. In late December 2025 a clip about alleged Minnesota fraud went viral, the agency reacted to the virality, and within weeks it had suspended 6,900 Minnesota borrowers tied to roughly 7,900 PPP and EIDL loans worth about $400 million, then announced it was expanding nationwide. The SBA spokesperson framed the expansion as part of a broader zero-tolerance policy on fraud. Zero tolerance, triggered by an algorithm of a different kind: the one that decides what trends.
Think about the sequence, because the sequence is the whole farce. A video goes viral. The agency suspends thousands of people. Then it hires Palantir. The surveillance machine was not the thing that found the fraud. The viral video was. Palantir arrived after the suspensions, as the apparatus to justify and industrialize a dragnet that a social media clip had already kicked off. The order of operations is exactly backwards from how a careful enforcement program works, where you investigate first and act second. Here the action came first, the virality drove it, and the $300,000 Bootcamp is the part where the agency goes back and builds the machine that was supposed to have found the problem in the first place.
The Math That Makes It Theater
Now hold the surveillance budget up against the recovery reality, because that is where the absurdity finishes itself. The SBA recovers roughly two cents on every dollar it charges off in COVID loan fraud. That is the agency's own performance on money it is already chasing. So the economic case for the Palantir pilot is that a $300,000 introductory contract, scaling into who-knows-what once the Bootcamp graduates, will help recover fractions of pennies on a dollar that left the building in 2020. The cost of the surveillance is fixed and real. The recovery it enables is theoretical and tiny. The only thing that scales cleanly here is the data collection.
And that is the tell. When an agency spends real money on surveillance to chase money it cannot recover, the surveillance is the point, not the recovery. The dossiers get built whether or not the dollars come back. The risk scores get assigned, the profiles get assembled, the unified data environment gets populated with hundreds of thousands of borrowers, and the fact that the underlying program will claw back almost nothing does not slow any of it down, because none of that infrastructure has to justify itself by recovering money. It justifies itself by existing. The $300,000 buys the foothold. What it builds is permanent.
Who The Risk Score Actually Catches
The grimmest part is the same grim part that runs through every story on this site. A risk-scoring dragnet does not catch the professional. The industrial fraud rings that ran application mills, the synthetic identities, the dissolved shells, they were built to be unfindable, and a Foundry case file does not resurrect a borrower who stopped existing in 2022. What the machine catches is the findable, the on-the-grid, the small operator who took a real loan during a real shutdown, used it the way the program instructed, got forgiveness years ago, and is now sitting in a dataset being ranked against criteria they will never see, by a vendor they never heard of, on a contract that calls the whole thing a Bootcamp.
That is the through-line. The algorithmic suspension of 111,620 California borrowers flagged first and demanded a negative be proven later. The expanded administrative courtroom let the agency judge and collect under its own roof. The Palantir contract is the engine room for all of it: a $300,000 pilot for a machine designed to turn an entire borrower population into a ranked list of suspects, kicked off by a viral video, aimed at money that comes back at two cents on the dollar, and filed under a word that belongs at summer camp. The fraudsters are gone. The Bootcamp is for everyone who stayed.