ChatGPT's new image model makes document forgery much easier

Mar 31, 2025

GPT-4o prompted by THE DECODER

Key Points

OpenAI's new GPT-4o image model can generate realistic-looking fake document photos, such as bank transfers, doctor's letters, or certificates, in just seconds.
The ease with which convincing document forgeries can now be created by virtually anyone may overwhelm security checks at companies and authorities due to an increased volume of forgeries.
While OpenAI includes special metadata in all images generated by GPT-4o to identify them as AI-generated, this safeguard is only effective if it is actively checked and can be bypassed by knowledgeable users.

OpenAI's latest image model can create convincing fake documents in seconds, potentially overwhelming existing security systems by giving anyone access to sophisticated forgery tools.

X user "God of Prompt" shared several examples highlighting the model's capabilities, showing how GPT-4o can generate deceptively realistic photos of flight cancellations, bank transfers, university degrees, or medical prescriptions.

Screenshot of a fake flight cancellation. — A generated fake flight cancellation document. | Image: via God of Prompt

While document forgery has always been possible using traditional image editing software, AI automation fundamentally changes the landscape. Instead of requiring hours of expert work, convincing fake documents can now be created by anyone in seconds.

An AI-generated medical note showing illness symptoms for our editorial bot "Botti." | Image: GPT-4o prompted by THE DECODER

Security systems face new challenges

The potential volume of AI-generated forgeries could strain existing verification systems. This becomes particularly concerning in situations where organizations perform only cursory checks, such as small compensation claims, rental payment histories, or initial job application screenings.

Current control mechanisms at companies and government agencies might fail under the sheer volume of potential counterfeits. Even with consistent detection rates, the increased number of fake documents makes it more likely that some will evade detection.

A realistic-looking restaurant receipt showing itemized charges from "The Decoder Bar", including food and drink items, tax, and a total amount. The receipt appears to be photographed on a surface, complete with realistic shadows and creases. — A generated receipt from "The Decoder Bar" demonstrates how easily payment evidence can be fabricated | Picture: GPT-4p prompted by THE DECODER

OpenAI has built in several safety features to prevent misuse. Every image created by GPT-4o comes with special C2PA metadata that marks it as AI-generated. The catch is that anyone checking these images needs to use a specific verification tool to spot this metadata. Plus, savvy users might find ways to strip out these digital markers.

A computer interface showing C2PA metadata verification results. The screen displays technical details about an image's origin, creation date, and AI generation markers. — A screenshot showing how specialized tools can read C2PA metadata to verify an image's origin. | Image: Screenshot from THE DECODER

To help with verification, OpenAI says it has also developed its own internal tool for tracking if images were generated with GPT-4o, though it isn't clear who uses this tool when and for what.

In its GPT-4o system card, OpenAI states it aims to "maximize helpfulness and creative freedom for our users while minimizing harm." The company acknowledges that safety remains an ongoing process, with guidelines evolving based on real-world usage.

AI News Without the Hype – Curated by Humans

As a THE DECODER subscriber, you get ad-free reading, our weekly AI newsletter, the exclusive "AI Radar" Frontier Report 6× per year, access to comments, and our complete archive.

Source: via X