I’ve written recently about how AI can empower compliance teams to do their work.The flip-side is that fraudsters can also leverage AI.
In an ominous development, fraudsters are now using generative AI techniques including deepfakes, short for “deep learning fakes,” to create counterfeit documents, from bank statements to driver’s licenses to passports. Generative AI techniques enable fraudsters to convincingly manipulate the images and texts of key loan documents, making detection increasingly difficult and posing a significant threat to individuals and lenders alike.
FinCEN Report: Surge in AI-Driven Fraud Schemes
In November 2024, the U.S. Department of the Treasury’s Financial Crimes Enforcement Network (FinCEN) issued a report that fraud schemes involving generative artificial intelligence tools are on the rise. Specifically, “beginning in 2023 and continuing in 2024, FinCEN has observed an increase in suspicious activity reported by financial institutions describing the suspected use of deepfake media in fraud schemes targeting their institutions and customers.” FinCEN specifically reported on activities by criminals altering or creating fraudulent documents to circumvent verification methods.
FinCEN reports that fraudsters have used GenAI to falsify identity documents and passports, documents, photographs, and videos to get around know-your-customer requirements and verification checks.
Deepfakes Can Skirt Traditional Fraud Tools
This emergence of deepfakes is entirely unsurprising. GenAI tools enable fraudsters to create highly realistic loan documents in less time, with less money, and using fewer resources.
As one example, a fraudster can now subtly alter a photograph on a driver’s license replacing the image of one person with another. Even more concerning is the ability to manipulate the text itself, changing dates of birth, addresses, or other crucial information without leaving obvious traces of tampering.
Unlike traditional methods of counterfeiting, which often leave telltale signs of alteration, these deepfakes can be remarkably seamless. The manipulation involves altering the image, called “inpainting”, which leaves the surrounding document untouched. This means that the overall appearance of the identification document, including holograms, microprinting, and other security features, can remain intact, making it appear genuine at first glance. A visual inspection might fail to detect the subtle facial changes or text alterations, allowing fraudulent individuals to slip through the cracks.
How InformedIQ’s AI Identifies Deepfake Fraud
To combat these new fraud risks, lenders can’t do it alone. The challenge lies in detecting these sophisticated forgeries. Traditional methods of ID verification are no longer sufficient. Lenders need tools that can do more than evaluate isolated individual documents, but look at these documents in context, that can look at metadata, and be able to cross-reference information.
Here are examples of document-related fraud Informed’s AI tools can detect
- Contextual consistency among documents: Informed can check the logical consistency of information in a loan package: Does a provided paystub deposit also appear on a bank statement?
- Context within a document: Informed can assess whether a submitted document matches the providers’ formatting rules.
- Metadata: Informed can assess metadata in a .pdf such as what was edited by whom and compare it to known fraud template generators.
Leveraging Consortium Data for Robust Fraud Prevention
Perhaps, the most powerful tool against ‘deepfakes’ is consortium data. Fraudsters simply lack the insight to know how the data they are using has already interacted with the world, and the data residue it has left behind. To this end, Informed’s AI tools can extract new unique keys from transaction line items, document headers and footers, even visual content like barcodes and QR codes – and ensure that these keys are truly unique when compared to a consortium of millions of documents.
Further Insights into AI-Driven Fraud Solutions
For further background on Informed’s fraud solutions, check out these blogs:
