Re:Delimiters won’t save you from prompt injection
by graypegg on Thursday May 11, 2023 @06:56PM
(#35909454)
Yes totally. This really is no different than any other code injection vulnerability. Only allow symbols that you expect, and don't concatenate user input and logic unless the bounds between the two are guaranteed to be explicit.
Re:Delimiters won’t save you from prompt injection
by PeterisP on Saturday May 13, 2023 @09:36AM
(#35928230)
> don't concatenate user input and logic unless the bounds between the two are guaranteed to be explicit.
Well that's kind of the whole problem - LLM-based agents inherently work by literally concatenating logic with user input, and the bounds aren't guaranteed to be explicit. There is a discussion about finding a way to implement such bounds, but we don't have a good solution yet.
Re:Delimiters won’t save you from prompt injection
by vczf on Saturday May 13, 2023 @04:15AM
(#35926585)
There's no reason why delimiters need to be picked for human readability. Cryptographically random delimiters with adequate entropy seem to guard properly.
Re:Delimiters won’t save you from prompt injection
by tanseydavid on Thursday May 11, 2023 @08:30PM
(#35910313)
I was going to assert that the 'system role' provided by the API should prevent this problem if used properly.
But then I stumbled this recent information, which seems to say that the 'system role' is not quite behaving as intended or as you might expect from reading the docs.
Re:Delimiters won’t save you from prompt injection
by squeaky-clean on Saturday May 13, 2023 @03:19AM
(#35926341)
Have you managed to break recent versions of Bing Chat? I've tried some of the notable injection prompts and my own variants. Sometimes I can get it to begin with the jailbreak, but after 10-20 words it will erase the previous text and say something along the lines of "I cannot talk about that"
Re:Delimiters won’t save you from prompt injection
by codeflo on Saturday May 13, 2023 @03:33AM
(#35926409)
My reading of these events is that the jailbreak is successful, but some other, slower system detects it after the fact and deletes the message. Given the delay, I've seen speculation that there's a manual step involved.
Re:Delimiters won’t save you from prompt injection
by cmelbye on Saturday May 13, 2023 @01:10AM
(#35925775)
They didn’t promise that “system” fixes the problem. It is a prerequisite to a possible solution for the problem. Now that the model knows which messages are “system” and “user”, it can be trained to treat them differently.
Which is achieved how for an LLM?
Well that's kind of the whole problem - LLM-based agents inherently work by literally concatenating logic with user input, and the bounds aren't guaranteed to be explicit. There is a discussion about finding a way to implement such bounds, but we don't have a good solution yet.
See my other comment here: https://news.ycombinator.com/item?id=35926548
But then I stumbled this recent information, which seems to say that the 'system role' is not quite behaving as intended or as you might expect from reading the docs.
https://community.openai.com/t/the-system-role-how-it-influe...
https://simonwillison.net/2023/Apr/14/worst-that-can-happen/...
That thing where Bing redacts text after the point isn't going to protect your application against problems like those I describe in https://simonwillison.net/2023/Apr/14/worst-that-can-happen/