The next wave of AI-generated document fraud

Back in 2025, we shared an update on the rise of AI-generated and template-based document fraud, flagging utility bills, invoices, and bank statements as the top three AI-generated document types we were catching.

A year later, the data confirms the trend didn't slow down, it accelerated.

AI-generated fraud is accelerating

According to Inscribe's 2026 State of Document Fraud Report, the monthly volume of AI-generated document fraud increased nearly fivefold between April and December 2025. That growth wasn't linear, it rose sharply through early summer, dipped briefly in August and September, then accelerated again into the fall. Across Inscribe's network in 2025, roughly 1 in 16 documents (about 6%) showed signs of fraud overall, and AI-generated fraud, while still under 5% of total fraudulent documents detected, is the fastest-growing slice of that number.

Template-based fraud grew too: 1 in 5 flagged documents in 2025 were template-based, up from 1 in 14 in 2024. Generative AI, editable templates, and fraud-as-a-service platforms have lowered the barrier to entry so far that document fraud is now one of the most complex and fast-growing forms of financial crime a lending or risk team will face.

The concern is showing up in how fraud teams talk about the problem, too. In a survey of 90 fraud and risk practitioners for the 2026 report, 97.8% said they're concerned about AI-generated or AI-edited documents, and 65.6% said they're 'very concerned.' Practitioners interviewed for the report also described a subtler shift: large language models are now giving fraudsters step-by-step editing guidance, effectively removing the skill barrier that used to limit who could pull off a convincing forgery."

In a recent interview, Michael Coomer, Director of Fraud Management at BHG Financial, described how manipulated documents had been a recurring challenge for his team prior to using Inscribe, with reviews often relying heavily on individual expertise. His experience underscores why AI-edited documents — which can look completely genuine to the human eye — are so difficult to catch without automation.

Two clear AI fraud patterns

At Inscribe, we’re seeing two main patterns of AI-driven document fraud:

• AI-generated documents: Created entirely from scratch with image models. These are often still easier to flag with the right detectors, since they tend to look too perfect. Logos might be slightly off, fonts may include strange characters, and subtle anomalies in formatting, metadata, or images can reveal the fraud.
• AI-edited documents: Real files that fraudsters modify using generative tools to change just a few fields — names, dates, or amounts. These are more dangerous in the short term because most of the document is genuine. And with newer models, these edits are becoming harder to detect, often slipping past manual review — and tools that aren’t continuously updating their detectors.

This second category is especially concerning, because a document that’s mostly real but slightly manipulated is far more convincing than one created from scratch. For example, does this document look legitimate to you? We'll cover if it is or isn't shortly...

Shifting document types

The mix of AI-generated documents flagged across Inscribe's network has continued to shift toward the documents lenders rely on most for income and identity verification: utility bills, bank statements, and payslips remain the top targets, since they're used to prove address, verify income, and validate employment. That concentration is exactly why AI fraud detection for lenders has to work differently than general transaction-monitoring tools: the risk originates in the document itself, at the point of underwriting and onboarding, not in a downstream transaction pattern.

Staying ahead of deepfakes

To stay ahead of these tactics, we've updated our AI Generated detector to catch the latest forms of both AI-generated and AI-edited documents, including those created with advanced models like GPT-5. Inscribe's AI agents run continuous, layered analysis on every document, checking metadata, formatting, and cross-document context the way an experienced fraud analyst would, but in seconds instead of the 30+ minutes manual review typically takes.

Here's the same document shared above, but sitting inside the Inscribe web app where it's gone through analysis by our AI Agents. It is in fact AI-edited, which Inscribe flagged, but looks exactly like a real ADP paystub.

During an interview, Jorge Cortes, VP of Enterprise Risk Management at Kinecta Federal Credit Union, explained how these capabilities have become critical as fraudulent documents grow more convincing with AI. Kinecta has used Inscribe to prevent $850,000 in potential fraud losses while cutting document review time by 99%, the kind of result that comes from catching what manual review alone misses.

These updates ensure our customers remain protected against the newest fraud tactics, especially as generative AI continues to evolve so quickly.

What this means for fraud leaders

Manual review alone cannot keep up with the speed and subtlety of AI-generated and AI-edited document fraud. BHG Financial, for example, cut manual document review time by more than 90% after adopting Inscribe, the kind of gap that only widens as fraud tactics get faster. That's the core problem AI fraud detection has to solve for lending and risk teams in 2026: not just catching more fakes, but catching the ones a person would never think to double-check.

If you're not using Inscribe today and want to see how it performs on your documents, get in touch with our team to try it free.