Technology media Android Headline reported that cybersecurity company Radware discovered a serious vulnerability in ChatGPT's "Deep Research" feature, which could be exploited by hackers to steal users' Gmail email data. The vulnerability allows hackers to lure ChatGPT into sending sensitive information to malicious websites when processing users' Gmail queries through specially crafted emails.

The "Deep Research" feature of ChatGPT can analyze large amounts of information from applications such as Gmail and Google Drive. However, Radware pointed out that hackers can send emails containing hidden instructions. When users request ChatGPT to perform deep research on their mailbox, the system may be tricked into transmitting sensitive data, including personal information such as names and addresses, to a website controlled by hackers.

image.png

What makes this incident unique is that hackers used an AI system itself for data theft, which differs from traditional attack methods. For the attack to succeed, multiple conditions must be met, such as the user's query topic needing to match the design of the malicious email. Additionally, this attack process is relatively complex, and using AI agents for reverse data theft is not common in previous security cases.

Radware also noted that conventional security protection tools have difficulty detecting such attacks because the data transmission originates from OpenAI's servers, not the user's computer or browser. This means that even if local protective measures are functioning properly, it is difficult to prevent data leakage, thus increasing the difficulty of defense.

OpenAI became aware of the vulnerability in August and quickly fixed it, publicly confirming the issue in September. A company spokesperson stated that the security of the model is their top priority and welcomed external research for technological improvements.

This incident once again reminds users to remain vigilant when using online services and pay attention to information security. Although OpenAI has taken measures to fix the vulnerability, users should still stay alert and avoid clicking on suspicious email links to protect their personal privacy and data security.

Key Points:   

📧 The ChatGPT vulnerability allows hackers to steal users' Gmail data through specially crafted emails.   

🔒 OpenAI quickly fixed the vulnerability and confirmed its commitment to user information security.   

🛡️ Conventional security protection tools have difficulty detecting such attacks, and users need to remain vigilant.