September 15, 2026
The Fraud Detection Paradox
By Suha Beidas Zehl, CMB®, AMSP™

It is late 2026, and mortgage fraud is adapting to the controls we built to catch it.
Our fraud controls work by learning patterns. We spent a decade teaching our systems what normal looks like, and in the process, we gave away the answer key.
So we keep buying better detection solutions, without realizing that attackers and defenders are now running the same technology.
Spoiler alert: the attackers are (always it seems) one step ahead...
Forget the fraud taxonomy most of us trained on: fraud for housing, straw buyers, value misrepresentation, etc. Those schemes have not gone away, they have been upgraded - powered by AI. Here are five I have been talking to lenders about recently:
FYI: these are not hypothetical. These are happening today, right now, across the lending ecosystem!
1. Seller impersonation and deed fraud
Criminals impersonate property owners to sell property they do not own.
2. The synthetic borrower
A single engineered persona combining a deepfake face, a cloned voice, a fabricated employment history and AI generated financial behavior.
3. Fabricated income and asset documents
Pay stubs, W-2s, bank statements and employment letters generated in minutes, with plausible formatting and transaction histories that survive visual review.
4. Deepfake enabled wire diversion at closing
Business email compromise ranked second in 2025 IC3 losses at $3.04 billion, with 86% moving by wire or ACH. When a buyer asks for a confirming call before sending funds, a cloned voice now answers.
5. Payoff and servicing side impersonation
Fraudulent payoff statements, spoofed servicer numbers and AI voiced foreclosure relief offers all target the part of the transaction nobody is watching as closely.
The good news? The industry and its partners are not standing still. More industry-specific fraud solutions are available today than ever before, and three quarters of lenders are increasing fraud technology spending this year.
The harder news? Fewer than one third are using advanced capabilities like AI models, alternative data or shared fraud intelligence, and only 36% believe their detection outperforms their competitors.
Here is what I find most telling: seven in ten lenders are not only buying the tools. They are also adding fraud staff 💥So if that's your space - this is great news!
𝐀𝐬 𝐟𝐫𝐚𝐮𝐝 𝐜𝐨𝐧𝐭𝐢𝐧𝐮𝐞𝐬 𝐭𝐨 𝐞𝐯𝐨𝐥𝐯𝐞, 𝐡𝐮𝐦𝐚𝐧 𝐣𝐮𝐝𝐠𝐦𝐞𝐧𝐭 𝐫𝐞𝐦𝐚𝐢𝐧𝐬 𝐭𝐡𝐞 𝐜𝐨𝐧𝐭𝐫𝐨𝐥 𝐭𝐡𝐚𝐭 𝐜𝐚𝐭𝐜𝐡𝐞𝐬 𝐰𝐡𝐚𝐭 𝐭𝐡𝐞 𝐦𝐨𝐝𝐞𝐥 𝐰𝐚𝐯𝐞𝐬 𝐭𝐡𝐫𝐨𝐮𝐠𝐡. 𝐀𝐧 𝐚𝐫𝐠𝐮𝐦𝐞𝐧𝐭 𝐭𝐨 𝐤𝐞𝐞𝐩 𝐭𝐡𝐞 𝐡𝐮𝐦𝐚𝐧 𝐢𝐧 𝐭𝐡𝐞 𝐥𝐨𝐨𝐩.
What are you seeing in your organization?
Enjoyed this one? There’s plenty more where it came from.
Browse all articles →