Deepfakes are bad. Very bad. But something else is brewing, and its not looking good.
A new study from Inscribe, a document fraud detection company, found that AI-generated fraudulent documents increased nearly 500% between April and December of 2025.
The finding confirms what fraud investigators have feared – the technology that powers chatbots, voice clones and deepfakes has become a weapon of choice for fraudulent paystubs and bank statements.
But the report reveals that the use of AI is not really what people think.
“AI-Assisted Fraud” Is The Real Problem
Creating convincing paystubs used to require skill. Fraudsters needed Photoshop expertise, access to real document templates and enough knowledge of how to make the math add up for deductions.
With Gen-AI, fraudsters now use AI for the heavy lifting, completely replacing the need for any photoshop expertise whatsoever.
“The real issue right now is AI-assisted document fraud,” said Ronan Burke, CEO and Co-Founder of Inscribe. “It’s not AI magically generating perfect paystubs. It’s people using AI to smooth out the fonts, alignment, wording, and internal consistency so a forged document looks legitimate at a glance.”
Yes, Fraudsters Are Really Using AI For Paystubs And Bank Statements
With so many paystub generators online, are fraudsters really turning to AI? Its a question that I had been eager to find an answer for.
So I researched, and surprisingly (or perhaps not suprisingly) there are quite a few post on social media talking about it – in particular ChatGPT.
“Just a heads up,” one influencer post on Facebook, “ChatGPT make paystubs. Do with that information what you will.”
And online, there are websites that tout using Artificial Intelligence to generate paystubs and bank statements.
One site – Paystub creator sells their “AI Paystub Generator”.
There Is Some Fraud Magic With AI and Paystubs
AI may not be able to magically create a paystub, but it can magically “alter one” as I found out.
Using ChatGPT made it a snap.
I found a fake paystub on Scrib’d – where fraudsters love to upload their fake documents and just asked ChatGPT to change all of the information for me – put it in my name and give me an exorbitant income.
Even more incredible – it automatically updated all the withholdings for me.
In about 90 seconds – this is what I had. I guess the guardrails within ChatGPT don’t work!
It turns out, I wasn’t the only fraud fighter trying this technique – Marc Evans of the Las Vegas Police Department had stumbled upon the same technique using Google Gemini.
AI Fingerprints Are Hidden, But Tiny Traces Are There
Fraud detection teams at Inscribe.AI are hunting for digital evidence left behind by AI systems.Different AI models leave distinct traces in document metadata and in the generated images themselves.
For example, GPT 4 might leave one type of marker, while GPT 4.5 leaves another. And GPT 5 might leave another type of evidence altogether.
“The characteristics change,” Ronan Burke said. “We’ve seen that they’ve changed between model versions. That’s one thing we find when we look closely at the documents.”
Another thing Inscribe has found is evidence in an image’s metadata – sometimes it appears altered, and other times it’s missing altogether. Neither of those red flags is visible to underwriters, and that’s where AI becomes a killer app. It can spot those patterns instantly.
Everyone’s Concerned – 99% Of Organizations Are Bracing For More Document Fraud
Its no surprise that AI-document paystubs and bank statements are freaking out fraud leaders. In fact, in surveys, Inscribe reports that “97.78% of fraud leaders report concern about AI-generated or AI-edited documents, and 65.56% of respondents say they are very concerned about AI-enabled document fraud.”
It appears the only thing that is standing between the fraudsters and fraud fighter might be AI itself. Good machines to fight the bad machines.
And that might be are only hope.
