The following post is provided by Point Predictive, a MeridianLink® Marketplace partner.
By Frank McKenna
Co-Founder and Chief Fraud Strategist, Point Predictive
Faking a perfect bank statement or pay stub once took Photoshop skills and genuine templates. Now, tools like ChatGPT and Google Gemini let fraudsters create authentic-looking pay stubs in just 90 seconds. They’re so convincing most underwriters miss them.
But AI-generated documents are just the start of a much broader fraud threat.
On fraud-focused channels of the dark web and Telegram, mentions of AI and deepfakes jumped from 47,000 messages to over 350,000 between 2023 and 2024 alone. AI is quickly becoming the method of choice to defraud credit unions, banks, and lenders.
What AI fraud looks like in lending
To find out what AI fraud looks like in practice, you can merely browse Telegram or, in many cases, the open internet.
It’s there you will find that fraudsters are increasingly resorting to AI software with names like “CPN Wizard” that can fully automate the creation of synthetic identities. By choosing the right stolen Social Security numbers to pair with their fabricated identities and then populating public records, fraudsters can make them virtually undetectable.
If they need a perfect driver’s license to match those synthetic profiles, they can simply head over to OnlyFake.Org, where they can purchase an AI-generated identity card to do the job.
Then there are deepfake videos. Fraud forums are filled with AI-generated clips that show people rotating their heads, blinking, and smiling on command. These videos are designed to fool the liveness checks that lenders and dealerships use to verify borrowers during “Know Your Customer” processes.
Why traditional defenses are falling behind
The auto lending industry faces an estimated $10.4 billion in fraud exposure in 2026. That’s a nearly five-fold increase since 2010. With AI, the losses will only get worse. A single fraud operator who once created a few fake identities a month can now produce dozens.
Here’s the core problem: most fraud prevention tools were built to catch humans making mistakes. They check whether a driver’s license looks right, whether a paystub has the correct formatting, and whether a selfie matches a photo ID. AI-generated fakes are designed to pass exactly those checks.
Credit bureau data alone can’t solve this either. A synthetic identity can have a real credit file built over months or years. Traditional red flag systems generate false positives at rates of 1,000 to 1, creating decision fatigue that lets real fraud slip through.
To solve fraud, you must go beyond the credit bureau and look for the truth in other data.
Consortium data reveals truth
AI can generate a perfect paystub, create a convincing driver’s license, and produce a deepfake video that passes a liveness check at a lender. But there is one thing AI cannot fabricate, and that’s real lending history across hundreds of lenders and thousands of dealerships spanning years of real applications and loans.
That’s why proprietary consortium data has become the strongest defense against AI-powered fraud. Point Predictive’s repository holds over 90 billion risk data points from over 300 million historical loan applications spanning over $5 trillion in loan value.
When a synthetic identity applies for a loan, it may have a clean credit report and a credit score of 800. But against a consortium that tracks how identities, incomes, employers, and loan applications connect across the industry, the patterns become visible. Point Predictive has identified over 16,000 fake employers being used in fraud schemes and more than 250,000 frequent synthetic identity fraudsters actively targeting lenders.
The consortium approach works because fraud is never an isolated case. A fake employer tied to one application is connected to dozens more at other lenders. A synthetic identity used in Florida shows up again in Texas. These connections only appear when you can see across the entire industry.
The double win: Better fraud detection with less friction
The best part of this approach is what it does for honest borrowers. Most applicants are truthful when they fill out an application. They shouldn’t have to suffer through lengthy document requests because of the few who aren’t.
Lenders that use consortium data to verify income and employment can remove paystub requirements on low-risk loans. One lender that did this saw their capture rate triple, jumping from 13% to 40%. On average, lenders that remove paystub friction see their conversion rates double on low-risk loans.
This is the model that credit unions are already using. They are catching fraud that their previous systems missed while also speeding up approvals for their members.
Moving forward: From manual document review to data first
AI fraud is not a future threat. It’s here now, and it’s growing fast. The lenders who will come out ahead are the ones who stop relying on document-based verification or manual review as their primary defense and start using data that cannot be replicated.
The fraudsters are already using AI. The only question is whether your fraud controls are ready for it.
About Point Predictive
Point Predictive powers a new level of lending confidence and speed through artificial intelligence, powerful data insight from our proprietary data repository, and decades of risk management expertise. The company’s data and technology solutions quickly and accurately identify truthful and untruthful disclosures on loan applications. As a result, lenders can fund the majority of loans without requiring onerous documentation, such as paystubs, utility bills, or bank statements, improving funding rates while reducing early payment default losses. Subsequently, borrowers get loans faster, and lenders realize a more profitable bottom line. For more information, please visit pointpredictive.com.
Media Contact: Jordan Zane, [email protected]
