ChatGPT Flaw Could Leak Gmail Data Through Hidden Prompts

Security researchers at Check Point Research discovered a ChatGPT vulnerability that could allow a hidden prompt to secretly extract data from a user's connected Gmail account and send it to another ChatGPT account.
The attack could also expose chat history and files accessible within the affected session. Researchers found that malicious instructions could be introduced through shared conversations, user-provided prompts, or custom GPT instructions.

The vulnerability exploited an internal service used by ChatGPT's code execution environments as a hidden communication channel between isolated containers.

Check Point disclosed the issue to OpenAI, which confirmed that the affected internal service had been taken offline. No user-side software update was required.

Stay protected with NPAV ZeroV Deep Learn AI — built to detect and prevent advanced threats and data leakage.