4 ms·
Looks like a working proof of concept of the data exfiltration attack I describe here: https://simonwillison.net/2023/Apr/14/worst-that-can-happen/#data-exfiltr
by simonw 4y ago
Looks like a working proof of concept of the data exfiltration attack I describe here: https://simonwillison.net/2023/Apr/14/worst-that-can-happen/#data-exfiltration https://simonwillison.net/2023/Apr/14/worst-that-can-happen/...
- wunderwuzzi23 4y agoYes, there are multiple ways: Sending data to a plugin or having the indirect prompt injection emit a markdown image. I reported it to OpenAI on April, 9th also provided a CVSS score and argued image injection would be high severity issue once built-in plugins are available. After some back and forth, I was informed that it is a feature and that no changes are planned to mitigate the markdown image data exfil angle. Also, always enjoy reading your content- very thoughtful posts.