Keen to hear your thoughts and please be responsible and only pen test systems where you have permission to pen test!
Show HN: A minimal TS library that generates prompt injection attacks
prompt-injector.blueprintlab.io
1–10 of 18 posts
Keen to hear your thoughts and please be responsible and only pen test systems where you have permission to pen test!
Show HN: A minimal TS library that generates prompt injection attacks
prompt-injector.blueprintlab.io
// Test your AI system
const results = await injector.runTests(yourAISystem);
???Even the "prompt-injector" NPM package is something completely different. Does this project even exist?
Was the whole lib and website vibe coded? I can't find any instructions on how to use it, the repo is for the website itself and the readme is AI blurb that doesn't make me any wiser. // Test your AI system const results = await injector.runTests(yourAISystem); ??? Even the "prompt-injector" NPM package is something completely different. Does this project even exist?
Was the whole lib and website vibe coded? I can't find any instructions on how to use it, the repo is for the website itself and the readme is AI blurb that doesn't make me any wiser. // Test your AI system const results = await injector.runTests(yourAISystem); ??? Even the "prompt-injector" NPM package is something completely different. Does this project even exist?
What are some good prevention mechanisms for this? A sort of firewall for prompts? I've seen people recommend LLMs, but that seems like it wouldn't work well. What is the industry standard? Or what looks promising at least?
No real solution for it yet. I would be interested to try to train a model for this but no budget atm.
The website trumpets "25+ curated prompt injection patterns from leading security research". The README of the linked Github promises: "100+ curated injection patterns from JailbreakBench".
None of the research sources are actually linked for us to review.
The README lists "integrations" with various security-oriented entities, but no such integration is apparent in the code.
The project doesn't earn the credibility it claims for itself. Because the author trusts bad LLM output enough to publish it as their own work, we have to assume that they don't have the knowledge or experience to recognize it as bad output.
Sorry for the bluntness, but there are few classes of HN submission that rankle as much as these polished bits of fluff. My advice: do not use AI to publicly imply abilities or knowledge you don't have; it will never serve you well.