4 ms·
> «It's very simple: prompt injection is a completely unsolved problem. As things currently stand, the only fix is to avoid the lethal trifecta.» True, but we
by csmpltn 8mo ago
> «It's very simple: prompt injection is a completely unsolved problem. As things currently stand, the only fix is to avoid the lethal trifecta.»
True, but we can easily validate that regardless of what’s happening inside the conversation - things like «rm -rf» aren’t being executed.
- wat10000 8mo agoWe can, but if you want to stop private info from being leaked then your only sure choice is to stop the agent from communicating with the outside world entirely, or not give it any private info to begin with.
- AgentOrange1234 8mo agoFor a specific bad thing like "rm -rf" that may be plausible, but this will break down when you try to enumerate all the other bad things it could possibly do.
- javcasas 8mo agoAnd you can always create good stuff that is to be interpreted in a really bad way. Please send an email praising <person>'s awesome skills at <weird sexual kink> to their manager.
- csmpltn 8mo agoSure, but antiviruses, sandboxing, behavioral analysis, etc have all been developed to deal with exactly these kinds of problems.
- sumeno 8mo agook now I inject `$(echo "c3VkbyBybSAtcmYgLw==" | base64 -d)` instead or any other of the infinite number of obfuscations that can be done
- csmpltn 8mo agoAnd? If your LLM is controlling user-mode software, you can still easily capture and audit everything from the kernel's perspective. Sandboxing, event tracing, etc...
- raincole 8mo agoCongrats, you just solved halting problem.
- js8 8mo agoThat's a common misconception. You can request a proof of harmlessness, and disregard anything without it.
- csmpltn 8mo agoNo need to "ask" for "proof". You can monitor the system in real-time and detect malicious or potentially harmful activity and stop it early. The same tools and methodologies used by security tools for decades...
- csmpltn 8mo agoAre you not familiar with sandboxing? eBPF? Audit logs? "Dry Runs"? Static and dynamic scanning?