Post #1931635
2026-04-30 11:27 UTC
@dotjayne Almost. What I'm hearing is that you used an completely inappropriate tool for the task at hand.
Designing processes so that production can be taken down that easily—that sounds very wrong to me.
It literally doesn't matter if an AI agent or a human runs
'rm -Rf --no-preserve-root /' on that one box that kills your whole production.
Now giving a LLM (no matter how clever it sounds) access to the red button is an incredibly stupid thing on top.
Replies (1)
-
@wakame@tech.lgbt 2026-04-30 17:29
@yacc143 @dotjayne A few years ago, people could have argued that one needs to be especially careless and/or malicious to affect large-scale damage on a production system. In a way, I think it's very interesting that a entirely new danger category has emerged that is labelled "our own tooling". And since the AI tools are pushed by the companies themselves, it should in principle be harder to blame production problems on single employees. (In practice: "Why couldn't you magically forsee all problems?" and "Why didn't you check every line of that gigantic script?")