[cha1nsaw111]

i didn't find threads on this topic.
I'm interested in AI prompt injections, especially in the context of CLI tools such as Claude or Codex. I've gone through some of the basics but still feel like I'm missing a lot of information. Since models are now running in sandboxes, it has become harder to exploit them, but I'm sure there are still possible attack vectors. I'd love to chat about this if you have experience and would really appreciate any useful insights or resources you can share.

[DramaKing]

i didn't find threads on this topic.
I'm interested in AI prompt injections, especially in the context of CLI tools such as Claude or Codex. I've gone through some of the basics but still feel like I'm missing a lot of information. Since models are now running in sandboxes, it has become harder to exploit them, but I'm sure there are still possible attack vectors. I'd love to chat about this if you have experience and would really appreciate any useful insights or resources you can share.

There have only been a couple of prompt injections for freakin' RCE, and I'm not sure if they were for cloud models. I wouldn't try this. You haven't found any threads because this doesn't happen.

[cha1nsaw111]

There have only been a couple of prompt injections for freakin' RCE, and I'm not sure if they were for cloud models. I wouldn't try this. You haven't found any threads because this doesn't happen.

cli isn't cloud model. it runs in your terminal and and works with files on pc. i guess i haven't found threads because it's new niche of hacking most people don't know about

[DramaKing]

cli isn't cloud model. it runs in your terminal and and works with files on pc. i guess i haven't found threads because it's new niche of hacking most people don't know about

I don't know about that, just GitHub Copilot and Cursor AI.

[cha1nsaw111]

I don't know about that, just GitHub Copilot and Cursor AI.

so there is no info on that topic on this forum -> we can consider this new method