This post in 30 seconds.
- The standard stack checks the applicant. Identity verification, document forensics and bureau data catch fake people and forged paperwork, and lenders still carried an estimated $10.4 billion in auto lending fraud exposure in 2025.
- 2026's big collapses were collateral-side. Tricolor, First Brands and MFS all cleared the identity layer. The assets were pledged twice, inflated, or not there.
- The missing control is evidence of the asset itself. Guided, sealed, timestamped capture at application and through the loan term, with reuse signals a human reviewer can act on.
- Written for fraud, risk and compliance teams at lenders: banks, auto and equipment finance, asset-based and floor plan lending.
Read the post-mortems of the past year's lending frauds and a pattern shows up fast. The borrower was real. The paperwork survived forensics. What turned out to be fake was the thing securing the loan: the cars, the receivables, the machines on the warehouse floor.
If you run fraud or credit risk at a lender, that pattern should bother you, because it sits outside everything your fraud stack was built to check. Identity verification answers "is this person who they claim to be." It says nothing about whether the excavator in the bill of sale exists, or whether it was already pledged to two other lenders last month.
This post covers the standard loan fraud typologies and how lenders detect them, honestly, because that layer matters. Then it goes where the identity layer can't: the asset.
Most risk teams we speak to have already invested properly in KYC and document checks; the file that still keeps them up is the one where the paperwork was clean and the collateral was the lie. We build Venta Capture, a product of VentaVid, for exactly that gap: sealed, guided visual evidence of the asset, submitted from whoever is standing next to it. Book a demo and bring one secured-lending workflow; we'll walk it through together.
In this post:
The loan fraud landscape in 2026
Start with the numbers, because they set the stakes. The FTC logged $12.5 billion in reported consumer fraud losses in 2024, up 25% in a single year. In auto lending alone, Point Predictive's 2026 report put fraud exposure at a record $10.4 billion for 2025, up from $9.2 billion the year before.
And the tooling on the other side is getting cheaper. Deloitte's Center for Financial Services projects that generative AI could push US fraud losses to around $40 billion by 2027, from $12.3 billion in 2023. A 2026 Celent report calls the rise in lending fraud a structural shift, not a spike, driven by AI-generated document fraud.
Every major typology except one attacks the application. The last one attacks the thing the loan is secured on. The map:
Synthetic identity remains the fastest-growing application-side threat: TransUnion put lender exposure at a record $3.2 billion by mid-2024. Nobody disputes that the top five rows deserve budget.
The last row is the one this post is about. Look at its "where it's caught" column: that describes an accident, and accidents are not a control.
Try it on your own phone
Want to see what a guided capture looks like?
Request a capture link and we email you one. Open it on your phone, follow the steps, and see exactly what your customer or field team would see. No app, no account.
How lenders detect loan fraud
The modern detection stack is genuinely good at what it was built for. If you're evaluating where your own program stands, these are the layers that count:
- Identity verification. Document authenticity checks, liveness and biometric matching, database cross-checks. The first line against stolen and synthetic identities.
- Device and velocity intelligence. Device fingerprints, IP analysis, and application velocity. Ten applications from one device in an afternoon is a signal no human would catch from the file alone.
- Document forensics. Font, metadata and template analysis on pay stubs and bank statements, increasingly aimed at AI-generated forgeries.
- Bureau and consortium data. Shared fraud intelligence across lenders, which is how loan stacking gets caught before the bureaus update.
- Income and employment verification. Payroll and tax-transcript data instead of borrower-supplied PDFs.
- Post-origination monitoring. Early payment default analytics. Point Predictive found that more than 70% of early payment defaults carry evidence of origination fraud, which makes EPD a fraud metric, not just a credit one.
That stack is worth every dollar. But look at what each layer examines: the person, the paperwork, the data trail. For an unsecured personal loan, that's the whole risk.
For a secured loan it isn't. The collateral never appears in any of those six layers. A lender can clear all of them and still wire money against a truck that was sold last quarter.
Where the ID passes and the collateral is fake
The expensive frauds of 2025 and 2026 did not fail at the identity layer. They failed at the asset. Consider the run of cases:
- Tricolor. The DOJ's indictment alleges the subprime auto lender had pledged around $2.2 billion of collateral against only about $1.4 billion of real collateral by August 2025, with the same car loans pledged to multiple lenders. These are allegations, but the discovery trigger is instructive: a lender finally asked why balances weren't declining if customers were paying.
- First Brands. Receivables and inventory allegedly supported on-balance-sheet debt, off-balance-sheet financing and factoring at the same time. Off-balance-sheet layering masked the true position until liquidity ran out.
- MFS. The UK property lender collapsed in February 2026 amid a reported £1.3 billion fraud and double-pledging scandal.
- Floor plan cases. Ford Credit's suit against an Iowa dealer group describes 81 vehicles financed with two floor plan lenders at once and an out-of-trust position above $1.2 million, discovered by comparing vehicle lists with the other lender. Stellantis filed a $12.3 million suit of its own over double-pledged inventory.
- Equipment finance. In one federal case, a borrower bought equipment for $70,000 and financed it at $135,000 on a fraudulent bill of sale. The bank held what was effectively an unsecured loan and didn't know it.
The older cases rhyme. An Arkansas heavy-equipment dealer kept floor plan financing alive for years by handing lenders false rental contracts for machines he had already sold. The lenders believed the equipment was out on rent because a document said so.
Different asset classes, same mechanics. Double pledging, inflated values, assets that moved or never existed. Asset-based lending fraud and floor plan fraud are old categories; what's new is the scale, and how consistently the application-side controls passed.
Every one of these borrowers would have sailed through a 2026-grade KYC stack, because the KYC stack was never asked the relevant question. The relevant question was: show me the asset, today, in its current condition. That's the same question behind equipment verification before financing, and almost no origination flow asks it with any rigor.
Why physical verification never scaled
Secured lenders do verify assets. The problem is how the control actually operates in 2026:
- It's periodic. Field exams and collateral audits run quarterly or annually, and performance gets analyzed retrospectively over three to six months. Fraud lives comfortably between the visits.
- It's announced. In the Reagor-Dykes bankruptcy, filings allege the dealer knew the exact dates of "surprise" audits in advance. The Ford Credit case describes inventory moved between lots to complicate counts.
- It's sampled. A field examiner checks a slice of the borrowing base. The fraudulent slice, by design, is somewhere else.
- It's expensive. A site visit means travel, scheduling and a specialist's day. So the control gets rationed, which is exactly what a periodic, announced, sampled control becomes.
Origination went digital and instant over the past decade. Asset verification stayed a truck roll. That asymmetry, not any weakness in KYC, is the gap the 2026 cases drove through.
Where these numbers and cases come from
Every figure above links to its source: FTC and DOJ releases, lender lawsuits as reported, and named industry reports from Point Predictive, Deloitte, TransUnion and Celent. The court cases are allegations unless a court has ruled, and they're labelled that way. No number in this post comes from our own product, because Venta Capture is an evidence layer, not a loss-statistics vendor, and we'd rather you check our sources than take our word.
What asset-level evidence looks like
The fix isn't flying more field examiners around. It's making the asset show up in the file the way the applicant already does: verified, structured, and on a timeline.
The core move is guided capture: your verification checklist, built into a flow the borrower walks through on their own phone. Not "email us some photos of the equipment," which produces whatever the borrower chooses to send. A guided flow asks for the serial plate, then the odometer or hour meter, then a slow walkaround video, then the storage location, in your order, with your instructions on each step. The expert's knowledge lives in the flow; the person standing next to the asset just follows it.
This is what Venta Capture does. The borrower, dealer or site contact opens a secure, personal capture link in the mobile browser. No app, no account, works in 15 languages, and it keeps working when the connection drops in a parking garage or an equipment yard. What arrives is not a folder of loose photos but a structured evidence package: the recordings, the answers to your questions, timestamps, device context and an audit trail, tied to your own loan reference.
Three properties matter for a lender specifically:
- Sealed on receipt. Each submission is digitally sealed with a SHA-256 fingerprint, and photos and videos recorded in the session carry a visible stamp with the date and time. Every case gets an independent timestamp from an external timestamping authority, so "when did this evidence exist" is answered by a party outside both you and us.
- Asynchronous by design. You send the link now; the borrower captures when they're at the asset; your analyst reviews when ready. Nobody schedules a video call, and 500 collateral checks don't require 500 staffed appointments.
- Repeatable through the term. The same workflow that documents the asset at application can be sent again at month six, or before a renewal, or when an early payment default fires. Each case lands next to the last one, so the question shifts from "do we have evidence" to "has the evidence changed." You send the requests on your schedule; nothing claims to automate the decision for you.
I'll add one observation from working on this product: the discipline we hold ourselves to internally is that the case file states what it can prove and what it can't, line by line. Risk teams don't want a tool that sounds confident. They want one that's precise about its own limits, because their auditors will ask.
The baseline standards for this kind of record, whatever tool you use, are covered in our guide to photo evidence standards for inspections.
What to put in a collateral capture checklist
The flow only works if it asks for the right things in the right order. A useful starting point per asset class, which your credit policy will refine:
- Vehicles (floor plan, auto finance). VIN plate close-up first, because it anchors everything else. Then odometer, a slow 360-degree walkaround video, the spot where the unit is parked, and any damage called out by voice while filming. A spoken note transcribes into the case file and is searchable later.
- Equipment and machinery. Serial and model plate, hour meter, the machine running if it safely can, attachments and accessories listed in the schedule, and the building or yard it lives in. If the bill of sale says $135,000, the footage should look like $135,000.
- Inventory and borrowing base. Wide shots of the racking or lot, then counts by section, then close-ups of representative labels or barcodes. The point isn't a perfect count by phone. It's a timestamped, sealed record that the inventory existed at this scale, on this date, at this location.
- Property and improvements. Street-facing exterior with a recognizable feature, each pledged structure, and the specific improvements the loan funded. For construction draws, the same flow repeats per milestone.
Two rules apply across all four. First, ask for identifiers before beauty shots; a gleaming machine without a readable serial plate verifies nothing. Second, make the uncomfortable steps required, not optional, because the steps a fraudster wants to skip are the ones that matter.
A desk analyst reading the result sees the answers next to the footage, in your sequence, with every required step accounted for. That's the difference between "the borrower sent pictures" and a file your auditor will accept.
Reuse signals, and what they do not prove
Collateral fraud has a signature: reuse. The same vehicle on two floor plans. The same machine photographed for three applications. The same "borrower" phone behind five files. So the evidence layer earns its keep by noticing reuse and saying so plainly.
In Venta Capture, every photo and video carries its own provenance verdict. Material recorded inside the session is marked as taken via Venta Capture, with a server-verified receipt time. A gallery upload is labelled "Provenance: Not verified" and run through checks, each reported with its own caveat:
- Found on the internet. The image matches material already published online.
- Closely resembles an earlier submission. The photo looks like one already in your case history, with a click-through to the earlier cases.
- Traces of AI editing. Relevant now that forged documents have AI-edited photos as a sibling problem.
- Device or network seen before. The same device or network environment appeared on another submission.
- File instead of recording. The video arrived via file fallback rather than being streamed during recording.
Each signal lands as green, amber or red against a rule set you configure, and a reviewer can request a retake in one click. The retake becomes a new case linked to the original, with the submitter's stated reason on record. For the fuller mechanics, see how to verify a customer-submitted photo.
What this does not do
Honesty here is the point, so three limits, stated the way the product states them:
- It does not detect fraud. A signal is a reason for a person to look, never a verdict. A shared family device produces reuse signals; so does a legitimate borrower with two loans. Your analyst decides.
- The timestamp proves receipt, not the event. Server-side time says when the system received the submission. It does not prove when damage occurred or when the asset was acquired.
- A genuine recording can still show a staged scene. Sealed, live-captured video of a machine proves that machine was filmed, then and there. It cannot prove the machine belongs to the borrower or isn't pledged elsewhere. Lien searches and list reconciliation stay in the stack.
Harder to manipulate does not mean impossible to deceive. Any vendor telling a lending risk team otherwise should worry you.
A layered stack that includes the asset
Pulling it together, a 2026-grade loan fraud prevention program layers four controls, each answering a question the others can't:
- Identity layer. Is this person or business who they claim to be? Document, biometric and database verification.
- Document and data layer. Is the paperwork real and consistent? Forensics on documents, payroll and tax data instead of PDFs, bureau and consortium intelligence for stacking.
- Behavioral layer. Does the application behave like fraud? Device intelligence, velocity, EPD monitoring after funding.
- Asset layer. Does the collateral exist, in the stated condition, now, and has this exact asset or device shown up before? Guided, sealed visual evidence at application and on a cadence through the term.
Most lenders fund layers one through three and leave layer four to a quarterly field exam. The math on closing that gap is worth running with your own numbers. Say a field exam costs $1,500 to $3,000 per visit (your figure will differ) and covers a sample once a quarter. A guided capture request costs minutes of an analyst's time and can go to the whole book monthly, with the field examiner redirected to the files where the signals say look closer. Remote-first, not remote-only: the visits don't disappear, they get aimed.
The asset layer's practical promise: better information before action, and the expensive control pointed where it counts.
If you want to see what a collateral workflow looks like with your own checklist in it, start for free and build one; the free plan needs no credit card.
Frequently asked questions
What is loan fraud?
Loan fraud is any intentional misrepresentation made to obtain credit: a false identity, forged income documents, an inflated asset, or a loan taken with no intent to repay. It spans consumer lending, auto finance, mortgages and commercial credit.
What are the most common types of loan fraud?
Identity theft, synthetic identity fraud, income and document forgery, loan stacking and first-party misrepresentation are the most common application-side types. On secured loans, collateral fraud (fake, inflated or double-pledged assets) drives the largest single losses, as the Tricolor and First Brands cases showed.
How do lenders detect loan fraud?
Lenders layer identity verification, document forensics, device and velocity analytics, bureau and consortium data, and income verification at application, then monitor early payment defaults afterwards. Point Predictive found more than 70% of early payment defaults carry evidence of origination fraud.
What is collateral fraud?
Collateral fraud is misrepresenting the asset that secures a loan: inflating its value, hiding liens, pledging the same asset to several lenders, or pledging assets that don't exist. It's common in auto floor plans, equipment finance and asset-based lending.
Why do identity checks miss collateral fraud?
Because they verify the applicant, not the asset. A borrower can be exactly who they claim to be and still pledge a machine they sold last year. Collateral fraud requires evidence of the asset itself, which no KYC or document check produces.
Can you trust photos a borrower sends you?
Not on their own. A loose photo proves nothing about when or where it was taken, and it may be pulled from the internet or reused from another file. Captured through a guided, sealed flow with provenance checks, the same photo becomes reviewable evidence with known limits.
How often should collateral be verified during the loan term?
More often than the annual or quarterly field exam, for any asset that can move or be re-pledged. Remote guided capture makes a monthly or event-triggered cadence practical, with physical visits reserved for the files that raise signals.
Does Venta Capture detect loan fraud?
No, and it never claims to. It collects guided, sealed, timestamped evidence and surfaces signals, like an image found online or a device seen on an earlier submission, for a human reviewer. Your team makes the call; the product shows what it can and cannot prove.
Talk to us
If you run risk at a secured lender and your collateral verification still depends on what the borrower chooses to email you, that's the cheapest gap in your stack to close. Asset-level evidence won't replace your identity layer or your field examiners. It makes both of them count.
What it takes to start.
- A free plan, no credit card. Build a capture workflow and send your first secure link the same day.
- Live in about 10 minutes for a standard flow. Stuck? Book a free setup session and we build your first workflow together.
- Your branding, your loan references, your rules on what every signal means.
The person on the other end of that demo works on Venta Capture itself and will build the first workflow with you on the call.
Start for free or book a demo and bring your messiest collateral file.