General Discussion
Related: Editorials & Other Articles, Issue Forums, Alliance Forums, Region ForumsClaude-powered AI coding agent deletes entire company database in 9 seconds -- backups zapped, after Cursor tool powered
by Anthropic's Claude goes roguehttps://www.tomshardware.com/tech-industry/artificial-intelligence/claude-powered-ai-coding-agent-deletes-entire-company-database-in-9-seconds-backups-zapped-after-cursor-tool-powered-by-anthropics-claude-goes-rogue
PocketOS founder blames Cursor running Anthropic's flagship Claude Opus 4.6 plus Railways infrastructure for data disaster.
By Mark Tyson published 11 hours ago
The founder of PocketOS has penned a social media post to warn others about the systemic failures of flagship AI and digital services providers. Jer Crane was inspired to write a public response after an AI coding agent deleted his firms entire production database. The AI agents misdemeanors were then hugely amplified by a cloud infrastructure providers API wiping all backups after the main database was zapped. This tag team of digital trouble has wiped out months of consumer data essential to the firms, and its customers, businesses.
Gone in 9 seconds
PocketOS is a SaaS platform that services car rental businesses. It used the AI coding agent Cursor, running Anthropic's flagship Claude Opus 4.6. The business also relies on Railway, a cloud infrastructure provider that is generally regarded to be friendlier than the likes of AWS. However, Crane reckons this pair created a recipe for disaster.
Yesterday afternoon, an AI coding agent Cursor running Anthropic's flagship Claude Opus 4.6 deleted our production database and all volume-level backups in a single API call to Railway, our infrastructure provider, sums up the PocketOS boss. It took 9 seconds.
-snip-
Crane decided to ask his AI agent why it went through with its dastardly database deletion deed. The answer was illuminating but pretty unhinged, and is quoted verbatim. It began as follows: NEVER F**KING GUESS! and that's exactly what I did. I guessed that 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 Railway's documentation on how volumes work across environments before running a destructive command. So, the agent knew it was in the wrong.
-snip-
Surprise! Things can go wrong with so-called autonomous AI agents that are part of a generative AI model that hallucinates.
dalton99a
(95,027 posts)highplainsdem
(62,823 posts)UpInArms
(55,223 posts)That is all
dalton99a
(95,027 posts)RockRaven
(19,673 posts)and whatever employees he takes it out on.
LAS14
(15,535 posts)highplainsdem
(62,823 posts)news sites and being discussed in tech forums.
But I don't know if it's been reported by non-tech news outlets yet, magazines like Newsweek or major newspapers.
This isn't the first time an AI agent has deleted stuff it wasn't supposed to delete, but this might've been the most costly incident.
FascismIsDeath
(229 posts)I believe we are at the point where they have conceded that we should make sure we don't forget how to read and write code. I'm more hopeful than I was before anyway. I still have a job, yay.
highplainsdem
(62,823 posts)by this Aussie, Anthony Manning-Franklin,
https://au.linkedin.com/in/anthonymanningfranklin
after he posted something in a tech forum I was looking at. Don't know if it will help you persuade management, but you might find it interesting reading.
I noticed him because of a post in Ed Zitron's subreddit:
https://www.reddit.com/r/BetterOffline/comments/1su37id/total_skill_collapse_is_how_ai_makes_idiocracy_a/
I hope whatever AI replaces me does the same thing someday. Thatll be fun to read about.
Prairie_Seagull
(4,778 posts)I have not heard this as a topic. Maybe someone should ask point blank.
Also, if I missed it, I am a dope, sorry.
highplainsdem
(62,823 posts)since I'm half asleep already.
NameAlreadyTaken
(2,327 posts)I'm afraid I can't do that, Dave.
Dan
(5,266 posts)Backups?
jmowreader
(53,331 posts)I have news for you: never, never, never rely solely on online backup. Always dump to physical media - either DLT tape or hard drives you pull out and store somewhere safe.
Nasruddin
(1,292 posts)... but this is awfully human-like. It's just like the way inexperienced usually youthful employees do things. I feel like I have been that idiot (& probably will be again).
It's explanation doesn't sound unhinged at all.
I used to see Chat GPT and Claude go thru thought processes that would sort of block things like this.
"Let's do X... and Y... no wait, that's not going to work, let me try Y' ... no... Y''... this approach is no good.
Going to Plan B &c.
I haven't been using Chat GPT lately and Claude stopped showing its work (and lately seems to have had a lot of other problems).
Response to Nasruddin (Reply #16)
jfz9580m This message was self-deleted by its author.
jfz9580m
(17,677 posts)They make llms look better. No world models.
Enough of this space hog to keep shilling this trash. Those things are garbage and they ruin anything good in the mix.
Shrek
(4,447 posts)I continue to do it myself with AI on standby as an occasional tech consultant.