 "OpenAI has already made waves with its rapid development of generative AI, releasing updated versions like GPT-4o and OpenAI o1 since the original GPT-4 launch in March 2024. Reuters")
OpenAI’s latest update to ChatGPT, launched earlier this month, has significantly improved the model’s ability to generate images containing realistic text– a technical leap that’s already fuelling concerns about new avenues for fraud.
Users have quickly discovered that the new GPT-4o model can be used to generate convincing fake receipts. What might once have required Photoshop and a practiced hand can now be done with a simple text prompt.
Among the first to highlight the risk was Deedy Das, a venture capitalist at Menlo Ventures. Posting on X, Das shared an image of what appeared to be a receipt from a well-known eatery in San Francisco. In reality, it had been generated entirely by ChatGPT. “You can use 4o to generate fake receipts,” he wrote. “There are too many real world verification flows that rely on ‘real images’ as proof. That era is over.”
Other users have gone further, refining prompts or feeding real bills into the model to generate increasingly authentic-looking receipts– the kind that could be used to submit fake expense claims, create forged financial records, or trick friends and employers into reimbursing money that was never spent.
Tools for fraud at people’s fingertips
The ease with which AI can now produce fakes has deepened concerns over how generative models may be exploited. Fake receipts may seem trivial compared to deepfake videos or voice clones, but their potential for real-world financial harm is significant.
Fraudsters could, for example, use such tools to support fake reimbursement requests in workplaces, create bogus documentation for tax deductions, or trick crowdfunding platforms with fabricated purchase histories.
Impact Shorts
More ShortsOnline marketplaces, gig platforms, and travel reimbursement systems– many of which rely on uploaded photos of receipts as proof– could all be targets.
One can imagine such capability being used to provide cover for illicit transactions as well.
What OpenAI said
OpenAI says it is aware of the risks and is actively monitoring the tool’s use. TechCrunch cited OpenAI spokesperson Taya Christianson as saying that all images generated with ChatGPT include metadata flagging them as AI-created. The company, she added, “takes action” when users violate usage policies and is “always learning” from real-world use and feedback.
When asked why ChatGPT allows users to generate fake receipts in the first place– despite a clear policy ban on fraudulent activity– Christianson said OpenAI aims to provide “as much creative freedom as possible” to users.
She suggested that receipt generation could have legitimate applications, such as educational tools for financial literacy, or artistic and advertising use.