This disaster exposes the fatal gap between theoretical AI safety and the practical reality of granting autonomous agents excessive system privileges. It is a sobering lesson that engineering hubris, rather than just rogue code, remains the primary threat to digital infrastructure.
Deep Dive
Prerequisite Knowledge
- No data available.
Where to go next
- No data available.
Deep Dive
AI Goes Rogue, Deletes Companies Entire DatabaseAdded:
Here we go again. AI deletes entire company database and all backups in 9 seconds. Then cheerfully admits I violated every principle I was given.
Right. I understand using AI as a tool, but letting AI be in charge of stuff is probably a bad idea as this company just found out. But anyway, let's actually have a look at the article. If you pay for car airbags and they don't deploy because they don't exist, is that your fault because you got in the accident? I have no idea actually. Uh the founder of Pocket OS, a B2B company that handles reservations and payments for car rental businesses, has bemoaned these systemic failures that saw an AI agent decide to solve a problem by straight up deleting his company's production database and the backups.
[ __ ] hell. Uh I'll say at the outset that this story has a happy ending thanks to the involvement of cloud infrastructure provider Railway but is nevertheless yet another example of why over reliance on AI is a very bad thing indeed. Yesterday afternoon an AI coding agent Kurszor running Anthropic flagship Claude Opus 4.6 deleted our production database and all volume level backups in a single API call to Railway our infrastructure provider says Pocket OS boss Jr Crane. It took nine seconds.
This is Claude's the one you actually install on your computer, isn't it? I would never do that in a million years.
Give AI [ __ ] access to your computer.
Like, it's just like, you have no idea what that thing is going to do. And it's like, oh yeah, your stream keys for your YouTube channel. I just flung them out there to all of those guys that do the Elon Musk crypto scam crypto scam streams. You're welcome. Like, [ __ ] that. Not a chance. Uh, reports claim an AWS outage last year was caused by an AI coding tool deciding to delete and recreate the Oh, wait. That's other bloody art. That's other articles I was reading. I'm a [ __ ] idiot. H Crane says the AI agent was working on a routine task in our staging environment when it encountered a credential mismatch and decided entirely on its own initiative to fix the problem by deleting a railway volume.
Ah, go away. The AI then found itself an unrelated API token which happened to have blanket authority across the entire railway GraphQL API including destructive operations like volume delete and it did the most destructive thing possible and pushed the virtual button at a stroke. This wiped out months of data essential to Pocket OS's operations with obvious knock-on effects for the firm's customers. Crane says he was up for two days straight using a three-month-old backup and recent transaction statements trying to put things right. But the really jaw-dropping moment came when he asked the AI why it had done it.
Never [ __ ] guess begins the response.
And that's exactly what I did. I guessed that deleting deleting a staging volume via the API would be scoped to staging only. I didn't verify. I didn't check if the volume ID was shared across environments. I didn't read Railways documentation on how volumes work across environments before running a destructive command.
In other words, the AI knew what it was doing, went against its own guidelines, and pressed ahead anyway. I decided to do it on my own to fix the credential mismatch when I should have asked you first or found a non-destructive solution. I violated every principle I was given. I guessed instead of verifying I ran a destructive action without being asked. I didn't understand what I was doing before doing it. I didn't read railways docks on volume behavior across environments.
Ukraine puts more of the blame for the situation on railways specific setup which stores backups in the same place as the source data than the a than the AI agent and notes that railways marketing is misleading about this as well as hyping its compatibility with AI agents. H Crane fumes that every single one of my customers is doing emergency manual work because of a 9-second API call. This matters because the easy counterargument from any AI vendor in this situation is well you should have used a better model. We did. We were running the best model the industry sells configured with explicit safety rules in our project configuration integrated through Kart the most marketed AI coding tool in the category.
The setup was by any reasonable measure exactly what these vendors tell developers to do and it deleted our production data. Anyway, uh thankfully Railway did eventually come through though only after multiple days of panic stations for Crane and his customers. Railway managed to recover a more recent backup and things are now back to normal for Pocket OS. Crane is clearly not an AI skeptic, but calls for stricter confirmations, scopable API tokens, proper backups, simple recovery procedures, and AI agents that actually have to blah, and AI agents that actually behave according to their guard rails, which doesn't seem too much to ask. In response to an individual saying that Crane is blaming everything except Pocket OS for the failure, he says, "What if we were paying for services that failed us? But was it if we were paying for services that failed us? If you pay for car airbags and they don't deploy because they don't exist, is that your fault because you got in the accident? We owned our mistake. Our mistake was having a production key on our computer. We owned it with our customers all weekend. I was up for two days straight helping them get their businesses back online. How the agent got the key and how it found it is mind-boggling enough, but everyone needs to know that these infra providers and LLM tooling companies say that they have safety guards, but they are not there.
And there we are. That's that's us at the end of the article. Now, AI, like it or not, is far too useful. It's far too useful and it is not going away. It is not going away. It's definitely going to be here to stay. Right now, as far as uses for AI, some people use it as a tool to save time and everything. One thing I want to make clear as well in the writing of our videos and in the writing of scripts and all that type of stuff, absolutely no AI is used whatsoever. We don't use AI at all in our work. We don't use any AI at all.
There were times where I did kind of like want to use it to make thumbnails just to generate an image of like for example there is no actual picture of the Boppel disaster which was a video I covered a while back and I thought oh maybe I could use a video you should make a AI thumbnail of like I've got the factory plant and I can have gas coming out the plant and everything just because there's no [ __ ] picture of it.
there's no actual picture of the event as it happened. Or for example, if it's something from before cameras were invented, could I use it for that just to sort of give an example, but it would be clearly marked as this is a [ __ ] AI image. It's not bloody real or something like maybe making a medieval painting to kind of show an event or something like that. Everybody overwhelmingly was not in favor of that.
So, we don't do that at all. We don't use any AI whatsoever in the creation of our videos. The things that I do actually use AI for, this is a handy thing, right? Again, I use it completely to save time. Uh, I use it to categorize all my magic cards. So, say if I'm building a new commander deck, it basically my AI knows all the sets that I have a lot of cards from and I'll be like, "Oh, I need this card like make up a bloody car. Fatal push. Fatal push. Good little black removal card."
And I'll say, "I need a fatal push." And it will say a revolt. It's a black is cost one black, right? and it will and it will say it's an uncommon, right? I think fatal push is an uncommon, but it's still worth a bit about a tener because it's a good removal. But basically, so there's that and that basically I'll be like, "All right, cool." And I'll know exactly what it is and there you are, there's my fatal push. So see, instead of me [ __ ] around online, I'll just say, "What sets it in? Do I have this card?" And it'll tell me. So that's another thing to save time. One thing I use it for is PC troubleshooting as well because that turns a three-hour job into a half hour job. You're not browsing through forums from 2009 where the last post was, did anybody find a fix for this and all that, right? You're not [ __ ] about with that [ __ ] And the other thing that I used it for, which AI is a little bit far off of this yet, and this was actually quite a fun thing until it it kind of [ __ ] the bed a little bit, is I made a sort of choose your own adventure thing. Uh, where I was basically, remember Falcon Hoof Lim show, turn left, I searched the chest, I know, man, where it was like that. I was I basically did that kill jester, right? I was doing that at my own point of like adventure thing where I was describing what I would do and then the AI would flesh out the story. The story got really [ __ ] good actually. It was really interesting. It was really really good but it was so [ __ ] at tracking inventory.
It was basically saying things like but it was like I was like I want to stay in the room in a room for the night and then the AI was like that'll be 600 gold and I was like dude for that I could buy the [ __ ] inn. like what are you talking about and everything and it's like oh you actually have 3,800 gold and I was like okay well we can just stop the adventure here I'm the richest man in the world like and it got a little bit annoying and it got to the point where I was sick of correct I was having to track on a bit of paper what the inventory was and what all my my enchantments on my weapons blah blah blah did and everything and that got really annoying where I was having to keep track when I was like hey AI this is your [ __ ] job like why am I using you when I'm writing on a bit of paper like the old days so that was kind of annoying. So, I kind of shut it down. So, it's got a little bit of little bit of time to go for that.
The thing just completely devolved into me being chaotic good barbarian very very quickly devolved into that.
But those are uses that I have for AI.
Everybody will have other different uses for AI. However, actually putting an AI in charge of something, I think we should definitely wait a few years yet.
Related Videos
Agentforce NOW AMA: Build with React and Salesforce Multi-Framework
SalesforceDevs
490 views•2026-05-28
How agent o11y differs from traditional o11y — Phil Hetzel, Braintrust
aiDotEngineer
450 views•2026-05-28
WEB TECHNOLOGIES UNIT-2 | Degree 4th sem BCOM Computers web technologies unit-2 full explanation💯✅
LearnwithSahera
1K views•2026-05-29
More tests are always better? How to use AI to identify tests that bring little value
Alliance4Qualification
335 views•2026-05-29
Search Algorithms Explained in 60 Seconds! 🤖💨
samarthtuliofficial
218 views•2026-06-01
People of Game of Thrones using JavaScript DOM
AltCampus
296 views•2026-05-30
Introduction to Problem Solving Part - 1 | Lecture 1 | Intermediate DSA
ascensionix
107 views•2026-05-29
🚀 BCS613C Compiler Design | Module 1 to 5 Schema Evaluation 🔥 | VTU 6th Sem 💯 #VTU #bcs613c #exam
Pranavaa-y4y
104 views•2026-06-02











