ChatGPT's new image generator is putting fresh pressure on a familiar trust problem: many systems still treat a photo as proof. According to TechCrunch, people are already using the 4o model's image tools to create fake restaurant receipts, including examples that imitate real-world paper, formatting, and wear.
The issue is not just that AI can make a receipt. It is that the output can be good enough to pass a quick human glance, while the remaining errors may be simple for a determined fraudster to correct.
Why fake receipts matter now
TechCrunch reports that ChatGPT unveiled a new image generator this month as part of its 4o model. The important change is its improved ability to generate text inside images, a weakness that often made earlier AI images easier to spot.
Receipts depend heavily on readable text. Store names, line items, totals, dates, and payment details are all visual evidence. If an image model can create that kind of text with fewer obvious artifacts, a fake receipt becomes more useful to someone trying to mislead a reimbursement system or verification process.
Deedy Das, described by TechCrunch as a prolific social media poster and VC, posted on X a photo of a fake receipt for a real San Francisco steakhouse that he said was created with 4o. His warning was direct:
You can use 4o to generate fake receipts. There are too many real world verification flows that rely on “real images” as proof. That era is over.
That point is the center of the concern. A receipt photo has often been treated as a practical shortcut. It is easy to request, easy to upload, and easy to review. But if a realistic-looking receipt can be generated on demand, the image alone becomes a much weaker signal.
What people were able to generate
The examples described by TechCrunch were not limited to clean, flat images. Others were able to produce similar receipts, including one with food or drink stains added to make the document appear more authentic.
Another example came from France, where a LinkedIn user posted a crinkled-up AI-generated receipt for a local restaurant chain. TechCrunch called that the most real-looking example it found.
These details matter because fraud detection often relies on visual intuition. A receipt that looks too perfect may raise suspicion. A receipt with wrinkles, stains, or uneven paper can feel more plausible because it resembles the kind of imperfect document someone might pull from a bag or pocket.
In plain terms, the image generator is not only producing text. It is helping create the surrounding cues that make a document feel lived-in and physical.
TechCrunch found flaws, but not enough comfort
TechCrunch tested 4o and was also able to generate a fake receipt for an Applebee’s in San Francisco. Its version had clear signs that it was not real.
One problem was formatting: the total used a comma instead of a period. Another was arithmetic: the math did not add up. TechCrunch noted that LLMs still struggle to do basic math, so that error was not especially surprising.
Those mistakes are important, but they do not fully solve the problem. TechCrunch points out that a fraudster could quickly correct some numbers with photo editing software or possibly with more precise prompts.
That creates a practical risk for expense workflows. If a fake receipt is almost convincing, the barrier to fraud may become less about creating the document and more about polishing a few obvious flaws.
The fraud risk is straightforward
The clearest potential abuse is fake reimbursement. TechCrunch says it is easy to imagine bad actors using this technology to get “reimbursed” for entirely fake expenses.
That risk extends from the way many organizations use receipts. A submitted image may be treated as evidence that a meal, purchase, or business expense happened. If the image can be generated without the transaction, the review process needs more than visual inspection.
The source article does not claim that every AI-generated receipt will work. It shows something narrower and still significant: the tools are improving, people are experimenting with them publicly, and some examples are realistic enough to force a rethink of image-based proof.
- Receipt images are easier to fake when an AI model can render legible text inside the image.
- Visual details can increase credibility when wrinkles, stains, or crumpled paper make the image seem physical.
- Obvious errors may not be durable safeguards if they can be fixed with editing software or better prompts.
OpenAI’s response
OpenAI spokesperson Taya Christianson told TechCrunch that all of its images include metadata indicating they were made by ChatGPT. Christianson also said OpenAI “takes action” when users violate its usage policies and that it is “always learning” from real-world use and feedback.
TechCrunch then asked why ChatGPT allows fake receipts to be generated at all, and whether that fits with OpenAI’s usage policies, which ban fraud.
Christianson replied that OpenAI’s “goal is to give users as much creative freedom as possible.” She also said fake AI receipts could have non-fraud uses, including “teaching people about financial literacy,” creating original art, and making product ads.
That response shows the central tension. A tool that can create a fake receipt may be useful in harmless or creative contexts. The same capability can also support deception when the output is used as evidence of a real transaction.
For businesses and platforms, the lesson is simple: a receipt image can no longer be treated as strong proof by itself. ChatGPT’s 4o image generator has made that assumption harder to defend.