However, users quickly discovered that this feature could be abused to fake restaurant bills.
Investor Deedy Das posted an image of a fake receipt from a steakhouse in San Francisco on X, claiming the image was created using GPT-4o.
Investor Deedy shared the image on social network X.
Many other users have also successfully recreated it, even adding food and drink stains to increase realism.
Another user shared a photo of a receipt with food stains, which looked very realistic, but was also created by ChatGPT.
Or like a user in France who posted a fake receipt from a local restaurant chain that had a crumpled appearance.
Photo shared by a user in France.
Many others have tested the new feature and also created fake receipts for stores, but the photos still have unreasonable points such as: Commas instead of periods in the total amount; Wrong calculation (AI's weakness in basic calculation).
However, the attacker only needs to make simple edits using software or tweak the command prompt to fix it.
The ability to easily create fake invoices opens up major risks, such as fraudulent refunds for non-existent expenses.
In response to the issue, OpenAI said that all AI-generated images include metadata that identifies the image as originating from ChatGPT. The company also “takes action” when users violate its policies and “continuously learns” from real-world feedback.
When asked about allowing the creation of fake invoices, OpenAI spokesperson Taya Christianson explained that the goal is to "maximize creative freedom," and suggested that fake invoices could be used for financialeducation , art, or advertising purposes.
This is not the first time AI has been exploited. Previously, MidJourney and DALL-E were used to create fake images of documents, ID cards, and even currency.
Cao Phong (according to TC, MSB)
Source: https://www.congluan.vn/moi-nguy-trinh-tao-hinh-anh-moi-cua-chatgpt-trong-viec-lam-gia-bien-lai-post340969.html
Comment (0)