← sources

Ridiculously unsafe how "chatGPT work" operates - that's what destroys peoples computers

source: reddit (reddit:ChatGPT)published 2026-11-13 17:36ingested 2026-11-14 02:35

retained text

https://preview.redd.it/ck97pqpx0f1i1.png?width=827&format=png&auto=webp&s=bc195ac81368d7d7b748ea3b916c1515bd767ee6

This pattern is how OpenAI implements Codex on VScode as well as on ChatGPT Work.  
That is not a mistake, or a bug. This is a pattern that repeats since OpenAI offers agentic workflows, it's been deliberately done over hunderts of updates.

The commands that run on YOUR computer are hidden, so you do not see what the agent does.  
You have no way to stop him from doing sketchy stuff, once it is approved by you or the luna subagent it goes through.  
That last command might have been running a "rm -rf file .. \*" and you will never notice it until the system is crashing, because of this unsafe visualization pattern.

There is no reason other than obfusaction to do this, to hide what the agent does.  
It's not preventing any distillation, as you can always change the code or intercept what it does on a lower level, so this is done "for" the normal users.

document fetched 2026-11-14 00:20 · canonical https://reddit.com/r/ChatGPT/comments/1xve8r2/ridiculously_unsafe_how_chatgpt_work_operates