ChatGPT Image Generator Creates Realistic Fake Receipts

ChatGPT's Enhanced Image Generation and the Rise of Fake Receipts

Recently, ChatGPT introduced a new image generator as part of its 4o model, demonstrating a significant improvement in its ability to render text within images.

Potential for Fraudulent Activity

This advancement has already sparked concern, as individuals are utilizing the tool to create convincingly realistic fake restaurant receipts. This introduces another sophisticated element to the growing arsenal of AI deepfakes employed by those engaging in fraudulent activities.

Deedy Das, a prominent figure on social media and a venture capitalist, shared on X a fabricated receipt for a legitimate San Francisco steakhouse, reportedly generated using the 4o model.

Replication and Authenticity

Numerous others have successfully replicated similar results. Some have even incorporated details like food or beverage stains to enhance the perceived authenticity of the generated receipts.

A particularly convincing example was discovered in France, where a LinkedIn user posted an image of a crinkled, AI-generated receipt from a well-known local restaurant chain.

TechCrunch's Testing

TechCrunch conducted its own tests with 4o, successfully generating a fake receipt for an Applebee’s location in San Francisco.

However, the generated receipt contained a few noticeable flaws. The total amount used a comma instead of a period, and the arithmetic was incorrect. Large Language Models (LLMs) often struggle with even basic mathematical calculations, making this outcome unsurprising.

Ease of Manipulation

Despite these imperfections, a fraudster could readily correct these numerical errors using photo editing software or, potentially, through more refined prompting techniques.

Fraud Opportunities

The ease with which realistic fake receipts can now be created presents substantial opportunities for fraudulent schemes. It is conceivable that malicious actors could exploit this technology to falsely claim reimbursements for nonexistent expenses.

OpenAI's Response

Taya Christianson, a spokesperson for OpenAI, informed TechCrunch that all images generated by their systems include metadata identifying them as created by ChatGPT.

Christianson further stated that OpenAI actively addresses violations of its usage policies and continuously learns from real-world applications and user feedback.

Policy Justification

TechCrunch inquired about the rationale behind allowing the generation of fake receipts, questioning its alignment with OpenAI’s policies prohibiting fraud.

Christianson explained that OpenAI aims to provide users with maximum creative freedom. She suggested that such receipts could have legitimate uses, such as educational purposes related to financial literacy, as well as in the creation of original artwork and advertising materials.

Further Considerations

• The ability to generate realistic text within images raises concerns about the potential for misuse.
• While OpenAI includes metadata, its effectiveness in preventing fraud remains to be seen.
• Continuous monitoring and adaptation of usage policies will be crucial in mitigating the risks associated with this technology.