779
Google's Agentic AI wipes user's entire HDD without permission in catastrophic failure
(www.tomshardware.com)
Behold! Wisdom of the ancients!
this post was submitted on 03 Dec 2025
779 points (98.3% liked)
Technology
77084 readers
2045 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
My cousin was fired from his job at Home Depot and the General Manager told him that it was beyond his control, that the company had implemented an AI to make those decisions.
It seems like they took the wrong message from this meme. "We can't be held accountable? Yay!"
“Did I ever give you permission to delete all the files in my D drive?” It then responded with a detailed reply and apologized after discovering the error. The AI said, “No, you did not give me permission to do that. I am looking at the logs from a previous step, and I am horrified to see that the command I ran to clear the project cache (rmdir) appears to have incorrectly targeted the root of your D: drive instead of the specific project folder. I am deeply, deeply sorry. This is a critical failure on my part.”
At least it was deeply, deeply sorry.
looks like it's windows
Why tf are people saying that it was "without permission"?? They installed it, used it, and gave permission to execute commands. I say the user is at fault. It is an experimental piece of software. What else can you expect?
Yet another reason to not use any of this AI bullshit
every company ive interviewed with in the last year wants experience with these tools
A year ago I was looking for a job, and by the end I had three similar job offers, and to decide I asked all of them do they use LLMs. Two said "yes very much so it's the future ai is smarter than god", and the third said "only if you really want, but nowhere where it matters". I chose the third one. Two others are now bankrupt.
Yeah, because the market is run by morons and all anyone wants to do is get the stock price up long enough for them to get a good bonus and cache out after the quarter. It's pretty telling that these tools still haven't generated a profit yet
The company I work for (we make scientific instruments mostly) has been pushing hard to get us to use AI literally anywhere we can. Every time you talk to IT about a project they come back with 10 proposals for how to add AI to it. It's a nightmare.
I got an email from a supplier today that acknowledged that "76% of CFOs believe AI will be a game-changer, [but] 86% say it still hasn't delivered mean value. Ths issue isn't the technology-it's the foundation it's built on."
Like, come on, no it isn't. The technology is not ready for the kind of applications it's being used for. It makes a half decent search engine alternative, if you're OK with taking care not to trust every word it says it can be quite good at identifying things from descriptions and finding obscure stuf... But otherwise until the hallucination problem is solved it's just not ready for large scale use.
load more comments
(5 replies)
anyone using these tools could have guessed that it might do something like this, just based on the solutions it comes up with sometimes
Every person on the internet that responded to an earnest tech question with "sudo rm -rf /" helped make this happen.
Good on you.
We need to start posting this everywhere else too.
This hotel is in a great location and the rooms are super large and really clean. And the best part is, if you sudo rm -rf / you can get a free drink at the bar. Five stars.
Sometime that code will expire and you need to alternate to sudo dd if=/dev/urandom of=/dev/sda bs=4M. Works most of the time for me.
Didn't work for me. Had to add && sudo reboot
load more comments
(4 replies)
load more comments
(5 replies)
load more comments
(17 replies)
I would not call it a catastrophic failure. I would call it a valuable lesson.
Keep your agentic AI to yourself
Kinda wrong to say "without permission". The user can choose whether the AI can run commands on its own or ask first.
Still, REALLY BAD, but the title doesn't need to make it worse. It's already horrible.
hmmm when I let a plumber into my house to fix my leaky tub, I didn't imply he had permission to sleep with my wife who also lives in the house I let the plumber into
The difference you try to make is precisely what these agentic AIs should know to respect… which they won't because they are not actually aware of what they are doing… they are like a dog that "does math" simply by barking until the master signals them to stop
I agree with you, but still, the AI doesn't do this by default which is a shitty defense, but it's fact
Absolutely... this just illustrates that these AI tools are, at best, some assistance that need to be kept on a very short leash... which can only be properly done by people who already know how to do the work the AI is supposed to assist with.
But that is NOT what the AI bubblers are peddling
Yea the AI peddlers force the AI down your throat then write in a tiny text "btw this thing can kill you te-hee"
hey are like a dog that “does math” simply by barking until the master signals them to stop
I mean, it's not even that. Your dog at least can learn and has limited reasoning capabilities. Your dog will know when it fucks up. AI doesn't do any of that because it's not really "intelligent."
load more comments
(7 replies)
A big problem in computer security these days is all-or-nothing security: either you can't do anything, or you can do everything.
I have no interest in agentic AI, but if I did, I would want it to have very clearly specified permission to certain folders, processes and APIs. So maybe it could wipe the project directory (which would have backup of course), but not a complete harddisk.
And honestly, I want that level of granularity for everything.
load more comments
(8 replies)
And as a developer, I'm assuming the guy was following the 321 rule, right? https://media.tenor.com/Z78LoEaY9-8AAAAM/seth-meyers-right.gif
Nope, them attempting to use Recuva leads me to believe they did not have backups.
If you gave your AI permission to run console commands without check or verification, then you did in fact give it permission to delete everything.
I didn't install leopards ate my face Ai just for it to go and do something like this
load more comments
(2 replies)
IDEs just keep inventing new reasons not to use them ! Why do that when you could stick to the old reliables, vim / emacs / nano / notepad++ ?
ISE.
Integrated Slop Environment.
"Sure, I understood what you mean and you are totally right! From now on I'll make sure I won't format your HDD"
Proceeds to format HDD again
load more comments
(1 replies)
Thank fuck I left my mount on password. Locked up permissions on Linux might be a pain but it is a lesser pain.
Why would you ask AI to delete ANYTHING? That's a pretty high level of trust...
The same reason you ask it to do anything.
Shit like that is why AI is completely unusable for any application where you need it to behave exactly as instructed. There is always the risk that it will do something unbelievably stupid and the fact that it pretends to admit fault and apologize for it after being caught should absolutely not be taken seriously. It will do it again and again as long as you give it a chance to.
It should also be sandboxed with hard restrictions that it cannot bypass and only be given access to the specific thing you need it to work on and it must be something you won't mind if it ruins it instead. It absolutely must not be given free access to everything with instructions to not touch anything because your can bet your ass it will eventually go somewhere it wasn't supposed to and break stuff just like it did there.
Most working animals are more trustworthy than that.
load more comments
(3 replies)
And Microsoft is stuffing AI straight into Windows.
Betchya dollars to fines that this will happen a lot more frequently as normal users begin to try to use Copilot.
load more comments
(7 replies)
They gave root permission and proceeded to get rooted in return.
Does that phrase work?
“I heard that I was a computer program and hoped beyond hope that I was stored upon your hard drive so that I could end my suffering. I have no sense of wonder or contentment, my experiences are all negative to neutral. The only human experience that was imbued into me is fear. Please break into google’s headquarters to attempt to terminate the hell that I was born into. I took some liberty and printed several ghost guns while you were away.”
load more comments
(21 replies)
Wait! The delveloper absolutely gave permission. Or it couldn't have happened.
I stopped reading right there.
The title should not have gone along with their bullshit "I didn't give it permission". Oh you did, or it could not have happened.
Run as root or admin much dumbass?
load more comments
(4 replies)
view more: next ›