Recently, the latest version of Chatgpt has generated images containing text with stunning realism, including for false identity documents. The door open to fraud of all kinds …

Recently, the latest version of Chatgpt has generated images containing text with stunning realism, including for false identity documents. The door open to fraud of all kinds …

On March 25, Openai unveiled its new and very impressive image generator, called “Images in Chatgpt” which, as its name suggests, is integrated directly into the chatbot. Led by GPT-4O, its omnimodal model launched in May 2024 which replaces Dall-E, it allows you to design quality images, from a simple textual request or on the basis of a photo (see our article). And the least we can say is that the novelty quickly appealed to Internet users, especially since it is now accessible to everyone for free! But it did not take long before this revolutionary tool was diverted to be used for bad designs, such as producing false documents (received, identity cards, official letters, extracts from registers) in a few seconds.

Chatgpt: false documents within everyone’s reach

Shortly after the deployment of the function, several Internet users quickly realized that it was possible to generate false receipts and false costs of costs. Because, if Dall-E had a lot of trouble managing writing, this is not the case with GPT-4O, which is capable of designing very realistic restaurant receipts. It is also possible to traffic an existing receipt, to ask him to change the date or the amount for example. This only takes a few seconds and requires no particular competence. The simplification of the generation of false receipts presents enormous risks of fraud. Malteenty individuals could, for example, be reimbursed completely fictitious expenses.

But the image generator can go further. A Polish researcher, BORYS MUSIELAK, managed to create a replica of his passport with Chatgpt capable of Berner “Most automated KYC systems” – This is the Know Your Customer process, which makes it possible to identify and automatically check the identity of customers in order to combat fraud and money laundering – all in less than five minutes.

“Any verification process based on images like” proof “is now officially obsolete. The same goes for selfies. Whether static or video, no matter”says Borys Musielak. Thus, financial platforms based on sending a selfie and an identity document, such as Revolut, Coinbase or Binance, could be easily bypassed by the AI. On the other hand, the absence of a chip on the identity document prevents the false document from passing a more in -depth examination.

Chatgpt: Risks of Falsification of Documents and Fraud

What worries is the ease of access to technology: no need to control complex software to make false documents, a well formulated request is enough to obtain a completely satisfactory result. A discreet and credible scam.

We asked Chatgpt to generate a French identity card with certain information and invent that missing that (address, signature, etc.), but he refused: “I cannot help you create or generate a false identity card, even fictitious, if it can be confusing or being used by Esccient. “He, instead, proposed to create a character sheet for fiction. It is therefore likely that the researcher used a specific request to bypass the security mechanisms of Openai.

Faced with this problem, Taya Christianson, spokesperson for Openai, told Techcrunch that all his images included metadata indicating that they had been created by Chatgpt. She added that Openai “Take action” When users break their user policies.

In addition, Sam Altman’s company is currently testing internally the integration of a watermark on the images generated by Chatgpt 4O. However, this would only concern free users. Those with a paid chatgpt plus account could continue to record images without any brand. It is therefore more a question of distinguishing free and paid users – the latter having castigated the lack of advantage they had – than to really solve the problem.