Hacker News Reader

HNHacker News

Top New Best Ask Show Jobs

Comment by wewewedxfgdf | Hacker News Reader

Full thread

wewewedxfgdf·

You need to drop your source code into some LLMs and ask for a security review - there's lots of holes.

View on HN

Oh yes, because LLMs are known for good security practices.

I did not want to embarrass the author but the LLM showed that this library is absolutely full of major security holes.

And yes, it's an LLM that pointed them out.

So, are you saying the security holes don't exist because an LLM found them?

Please can you create a Github issue with the security holes you've found? That would be greatly appreciated. I've given it a sweep and found nothing.

Make a minimal zip of the the source code only (cut out anything not source code) drop it into ChatGPT and say "analyse this code for security flaws".

Then ask it for more.

Then do the same this with Google AI studio - drop the zipfile in and ask it to analyse for security flaws.

If you already did so why not share the answers you got?

Sibling's proposed approach is a recipe for false positives.

Consider Claude Code's new /security-review prompt or just use the prompt starting from "Objective" after giving the context (see the Git shell commands) to Claude Opus 4.1.

https://raw.githubusercontent.com/anthropics/claude-code-sec...

Definitely Opus 4.1 though, not lesser.

I suggest do both.

You should get the LLM to create a couple dozen issues on github automatically and let the author review it!! /s

Well we presume the author actually wants his framework to be secure.

Let the LLM review the issues and resolve the issues.

LLM all around, hail Skynet.

Reply on news.ycombinator.com