A group of Chinese users had been using ChatGPT to create an AI tool for monitoring social media. For this, OpenAI banned the relevant accounts. The group had used ChatGPT to identify and edit errors in the tool’s code. OpenAI shared this information in a statement last Friday.
The group had asked ChatGPT to create a sales pitch for a program, which would help describe the program in an attractive way for marketing or selling purposes. Through the sales pitch, they aimed to highlight the program’s benefits and functionalities to attract users. The program was actually a surveillance tool that helped identify anti-China sentiment or protests on social media platforms (such as Facebook, X, YouTube, Instagram). The group’s intent was to send this information to the Chinese government.
OpenAI stated, ‘This network consisted of ChatGPT accounts primarily operating on Chinese business hours. The models were being prompted in Chinese, and ChatGPT’s tools were not being used as an automated program, but rather being personally utilized’.
The group had used the chatbot to verify or proofread the accuracy of the information they collected. This information was sent to the Chinese embassy and intelligence officials. These intelligence officials were monitoring movements in various countries, including the United States, Germany, United Kingdom.
Ben Nemo, the chief investigator at OpenAI, said that this is the first time OpenAI has encountered such an AI tool. By using AI models, what they are doing in other parts of the internet has come to our attention.
It seems that most of the code for the surveillance tool was based on an open-source version of the Meta’s LLaMA model. Additionally, the group used ChatGPT to write phishing emails on behalf of Chinese customers.
OpenAI also stated that they recently banned an account that was using ChatGPT to create social media posts against Chinese politicians and the refugee political scientist Kai Shaar in the United States. The same group had created reports in Spanish against the United States, which were published in major mainstream media outlets in Latin America. These were often published under the name of an individual or a Chinese company.
LND/SAKIB
