OpenAI's GPT-4o Model Reportedly Used to Generate Fake Receipts and Prescriptions
Menlo Ventures principal Deedy Das demonstrated that OpenAI's GPT-4o model can generate realistic fraudulent documents including fake receipts and controlled substance prescriptions. The model's advanced text-rendering capabilities make AI-generated images nearly indistinguishable from real ones, raising concerns about widespread document fraud and the obsolescence of image-based verification systems.
What happened
Menlo Ventures principal Deedy Das demonstrated that OpenAI's GPT-4o model can generate realistic images of fraudulent documents, including a fake receipt for a meal at a real San Francisco steakhouse with accurate item costs, subtotals, and tip calculations.
Das also showed that GPT-4o can generate fake prescriptions for controlled substances such as Zoloft. Another user added filters and artificial food stains to a generated fake receipt to make it appear more authentic for expense fraud purposes.
Reported impact
The realistic quality of GPT-4o's generated images could enable widespread document fraud including fake tax forms, bank cheques, counterfeit IDs, and birth certificates, according to the reports. Verification systems that rely on images as genuine proof are now obsolete, as the distinction between real and artificially generated images has become nearly impossible to discern.
Index analysis
GPT-4o distinguishes itself from previous AI image generators through its advanced ability to produce readable text within images that appears authentic. Security measures such as metadata tags and AI watermarking have proven to have simple workarounds.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.