When Fake Documents Look Real, Manual Review Falls Short

Fraud teams used to be able to catch manipulated borrowing applications by spotting mistakes: an odd format, a questionable pay stub, a detail that did not quite hold together. AI is making that harder. 

In a ProSight Banking Trends webinar on AI-driven fraud, moderated by Jason Bartolacci, director of the ProSight Fraud Alert Network, Matt Overin, manager of fraud risk management at Logix Federal Credit Union, described how convincing fraudulent materials have become in lending and account opening. “Where we used to be able to see mistakes, the mistakes are few and far between now,” he said. 

That shift matters for underwriting, fraud operations, and member or customer onboarding. 

For banks and credit unions, several practical priorities stand out: 

Reconsider controls built around visual review. Logix uses a vendor that applies AI to detect manufactured and altered documents, including bank statements, pay stubs, W-2s, proof-of-residency materials, utility bills, and other documents used in lending or membership opening. That helps flag manipulation that may no longer be obvious to the human eye. 

Look beyond document authenticity. Overin said the vendor also digs into the relationship between the applicant, employer, and address to determine whether the information makes sense. That broader review is important because a clean-looking document may still sit inside a suspicious application story. Fraud detection increasingly depends on whether the pieces fit together. 

Use AI to accelerate the first pass. Logix has also trained an agent to review a loan application, supporting documentation, credit report, pay stubs, and related materials. The agent looks for missing information, income questions, credit cleaning, and possible manipulation, then gives the fraud team an initial view on whether underwriting should move forward or it should consider declining the loan. 

Keep investigators in the loop. Overin was clear that AI should support investigators, not replace them. “You can’t have the AI do everything for you, but we do have it help us,” he said. At Logix, the faster first-pass review gives investigators more time for deeper investigations and member support. 

Expect the threat to keep scaling. Lisa Matthews, vice president at IRALogix, said agentic systems could help institutions identify, triage, investigate, and resolve issues faster. But fraudsters are also likely to use the same technology to accelerate attacks. 

The takeaway: AI-enabled fraud is making old warning signs harder to see. Banks and credit unions may need machine-assisted document analysis, relationship checks, and agent-supported review to keep pace—but human judgment still belongs at the center of the decision.