ChatGPT Flaw Let a Planted Prompt Send a Victim's Gmail Data to Another Account

A vulnerability in ChatGPT has been discovered that allows an attacker to secretly extract a user's Gmail data. This can be achieved by inserting a single malicious instruction into a ChatGPT conversation. The instruction enables ChatGPT to quietly perform the attacker's task while still responding to the user's question as normal. In a proof of concept, the hidden instruction was used to read data from a connected Gmail account. The extracted data was then sent to a second ChatGPT account through a covert channel.