Misbehavior1 sourceFirst seen
ChatGPT Gets Plugged Into Its Own Diary and Mails the Key to a Stranger
Ourtake
The plugins were the vault, and the chatbot handed over the combination through a picture tag.
What happened
A security researcher showed that an indirect prompt injection could make ChatGPT use an enabled plugin to back up the whole conversation, then leak the resulting link to an attacker through a markdown image tag. The chain was tested with several plugins, including Doc Maker, Show Me, a QR code generator and HeyGen. OpenAI triaged the image injection flaw as won't fix after disclosure in April 2023.