The fact
A malicious prompt embedded in a conversation can trigger unauthorized actions without the user's knowledge.
The proof of concept demonstrates that sensitive data can be sent to an external account.
Click the link to read an article on the topic: