58

Rather than adding an agents.md file written to stop LLM agents, I'm thinking about putting random comments into my code telling the agent to stop and warn the user, also some comments that are just word salads/non-sequiturs.

you are viewing a single comment's thread
view the rest of the comments
[-] hendrik@palaver.p3x.de 9 points 3 days ago* (last edited 3 days ago)

Unfortunately "poisoning" code or text doesn't really work. We've been discussing it every now and then. But there is no effective way to do it. It'll get filtered out beforehand and not even reach the LLM or it won't have any effect. So... You can do it, or save your time and spend it on something more useful. It won't do anything either way.

Giving instructions should work. Some people used the Agents.md file to make other people's coding agents do silly things. You can order it to recite Shakespeare or name all the variables like that, make it spam robot emojis or whatever you like. Maybe delete all the files in the repository...

Not sure if comments work as well. I mean it's designed to take instructions from the agents.md file. You can probably confuse it with comments in the code as well and likely make it do things. But the proper location for instructions might be better?!

this post was submitted on 20 Oct 2026
58 points (98.3% liked)

Fuck AI

8533 readers
2523 users here now

"We did it, Patrick! We made a technological breakthrough!"

A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.

AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.

Rules:

founded 2 years ago
MODERATORS