Using AI to Spot Fake IDs in Mortgage Lending

Mortgage Operations AI > Posts > Document OCR > Using AI to Spot Fake IDs in Mortgage Lending

Using AI to Spot Fake IDs in Mortgage Lending

Most lending teams do not take any measures to verify that a borrower’s ID is legitimate. Teams accept a picture of a driver’s license, often emailed from the borrower. They may look at the expiration date, but they usually take the information printed on the ID at face value.

The problem is multi-faceted here, and leaves mortgage lenders and banks open to fraud. First of all, pictures can be altered. I can take a picture of my ID and edit any of the data. I could change my name, DOB, signature, etc. With careful editing precision, I could also overlay a different picture onto my ID. Then I could export it from my photo editor as a jpeg and email it to the Loan Officer. That team would likely not notice if I was a good photo manipulator. The second issue is that there are companies that can produce fake IDs that look flawless. A fraudulent borrower could order a physical fake id, take a picture of it, and send a photo that is not at all manipulated. It’s a real photo, just of a fake physical ID.

There are several ways to do a more thorough job of verifying a driver’s license.

  • Ensure the borrower is using their portal to TAKE A PICTURE of their ID instead of uploading a pre-existing document. A document uploaded through the camera capabilities cannot be edited, plus the picture can usually contain meta data showing the longitude and latitude where the picture was taken and camera/phone identifying information.
  • AI document OCR functionality can pull the data from the front of the driver’s license, but there is still a problem of not knowing if it is fake
  • With the migration to REAL ID, all identification cards have a scan code on the back. Lenders can implement an API to read the data on that scan code. Perhaps AI can do this in the future. Then the lender can ensure the borrower info returned from the scan API matches the info on the front of the license.
  • If a public AI can be trained on all of the different state ID layouts, it could potentially identify an ID that has been tampered.
  • The photo from the front of the license should be used for to run some further verifications, particularly biometric checks, which I cover in another post.

If the scammers ever figure out how to game the scan code on the back, then we have a whole different issue to work through.