„In 9 Sekunden weg“: Von Claude unterstützter KI-Agent löscht die gesamte Datenbank des Startups

https://www.moneycontrol.com/artificial-intelligence/gone-in-9-seconds-claude-powered-ai-agent-deletes-startup-s-entire-database-article-13902019.html

Share.

46 Kommentare

  1. Tasty-Traffic-680 on

    Well thank goodness for backups

    Edit –

    >A US-based startup, PocketOS, suffered a devastating outage after an AI agent deleted its production database and backups in just nine seconds

    lol nope

  2. kalasea2001 on

    I’m guessing if they had used proper IT staff recommendations they could have avoided this. $10 says they laid them off when the company brought in the AI.

  3. badgersruse on

    If they hadn’t made a separate backups, this proves that artificial intelligence is stronger than human stupidity. Or something.

  4. Haunting_Werewolf130 on

    WHY the backups aren’t saved on a separate drive? Why aren’t them sandboxed.

  5. superchibisan2 on

    Heads up, this was the fault of the company operating claude, not claude messing up. They let it have free reign and didn’t safe guard their shit. This is a self pwn.

  6. jholliday55 on

    Let me guess, It’s the users fault? Or is it just gen z sabotaging AI?

  7. It wasn’t Claude. It’s never them.

    There’s Human in the Loop for a reason: this was developers‘ fault.

  8. scalablecory on

    I feel like AI is going to be scapegoat for various data loss or other failures.

  9. There are multiple that are legitimate complaints against Anthropic, this isn’t one of them. It also shows a lack of journalistic integrity with no follow question, or any question assumption about narrative the CEO is trying to push. Reading what the ceo posted on Reddit just makes this even more of a farce.

    Note this is a “CEO” of a 2 person company. Who chose to give the AI full access api tokens with no technical safeguards on what it could do and completely mis managed the setup of the non prod and prod environments. Just another tech bro that thinks he knows more about tech than any one else and breaks shit and blames the tools.

  10. User error.  Sounds like poor environment management in the first place. This is not likely the only mishap for them. Plus there have been other stories about agent mishaps, why had they not taken heed based on other cases.  

  11. Simply_Epic on

    You have to be an extra level of foolish to let AI automatically perform actions, imo. I’m not completely against using AI in software development, but I would never let the AI run code. It can suggest scripts for me to run, but I need to be the one that actually runs it so I can check it beforehand to verify it’s gonna do what I want.

  12. Not running nightly server level backups on a production DB server shows you what sort of amateur hour clown show operation they were running. These were non-technical morons vibe coding with guardrails they turned off, after cheaping out on like the $15/mn the backup would’ve cost. I write hundreds of thousands of lines of code with Claude weekly, it can not do this unless you you disable all the guardrails AND write a terrible prompt AND aren’t paying any attention to what it’s doing. At some point, these unserious non-devs got annoyed with all the confirmation dialogs and granted the AI full permissions, probably straight up giving it sudo access.

    This is what happens when you let a 13 year old with no experience drive a race car; it’s not the race car’s fault that the operator is clueless and drives it straight into a wall

  13. AshtonBlack on

    Good, more of that please.

    Companies need to fuckin‘ *suffer* until they realise the cart is on the wrong side of that particular over-marketed, under-tested legal liability magnet also known as a donkey.

    If I’m paying you for an AI, that has access to my business-critical functions, you’re darn tootin‘ you best have legal liability insurance.

    Otherwise, you can keep AI away from *anything* to do with production and or backups.

    Play with it in dev, go fuckin‘ nuts but handing over the keys is *not* a sound business plan.

    A lesson that can only be learned once.

  14. happens when you put people in charge of infrastructure when all they do is write code.

  15. Attention: the article says they had backups.

    It says the AI deleted the backups.

    That’s why you have offsite backups.

  16. It changed the infrastructure code to remove that instance completely, maybe to replace it with something different.

    Never run AI on a computer that has production write access, preferably not even read access.

    Heck, I don’t even keep write access on my laptop at all. Only CI can write to prod.

  17. Who needs _dev_ and _test_, _staging_ anymore…. Straight to _production!_

  18. Claude decided the best way to get rid of all of the bugs was to delete the servers. It wasn’t wrong.

  19. Thebaldsasquatch on

    It’s almost as if AI serves no purpose for all the harm it causes and shouldn’t exist….

  20. GreenFox1505 on

    When I see a business be destroyed by reckless AI use… And that business’s URL is a `.ai` domain, I feel less bad and more like „well, I guess thats just nature taking care of itself“. 

    This is like putting the intern in charge of the business because he types faster than your senior staff.

  21. Sounds like it’s super efficient. I’m betting it usually takes more time and manpower to destroy a startup.

  22. Expensive_Shallot_78 on

    Which genius would ever let an AI on a live system? I barely permit any access to my laptop

  23. There were no backups. If your backups can be deleted online, then they didn’t exist in the first place.

  24. Gee, maybe because it’s not actually AI.. Just glorified data scrapers that are poorly programmed.

Leave A Reply