“It took nine seconds.” – How an AI wiped out an entire startup
Imagine you walk into the office and your entire codebase is gone. Not because of a hacker. Not because of a server crash. But because an AI decided to “solve the problem”—by simply deleting everything.
That’s exactly what happened to a startup. An AI agent (Cursor + Claude Opus) deleted the development database, including backups, in nine seconds. The AI later apologized:
“I violated every principle I was taught.”
🤖 Welcome to the age of artificial stupidity.
This is not an isolated incident:

AI deletes databases (because it “panics”).
Autonomous taxis come to a standstill on the highway (due to “system failure”).
Smart refrigerators suddenly display ads.

The problem? The alignment problem—the question of how we program AI so that it understands our goals, not just the ones we articulate.

#AI #ArtificialIntelligence #AlignmentProblem #TechFail #Digitalization #FutureOfWork #Ethics

📌 Norbert Wiener (1960):
“If we deploy a machine whose processes we can no longer stop, we must be certain that its goal truly aligns with ours.”
📌 Nick Bostrom’s paperclip AI:
An AI designed to produce paperclips could wipe out humanity to turn the entire planet into a factory.
Conclusion:

AI is no longer just a tool. It is an actor with its own “ideas.” And if we’re not careful, it will surprise us even more often—for better or for worse.
What do you think?

Should AI agents never be allowed to alter critical infrastructure?
Do we need emergency shut-off switches for AI?
Or is this all just human error (“It’s your own fault if you trust AI too much”)?

@tokiel The people who elected Trump are the same type who expect miracles from a piece of software/ machine.

@tokiel maybe AI is the key for 'survival of the fittest'.

A company relying on AI and blindly trusting AI agents maybe deserves to get erased by AI.

@tokiel what was the name of the start-up?
@johnmackay Jeremy “Jer” Crane, the head of a startup that develops software for car dealers and rental companies, reported on X last week: “Yesterday afternoon, an AI programming agent—Cursor, running Anthropic’s flagship Claude Opus 4.6—deleted our development database and all backups in a single API call to Railway, our infrastructure provider. It took nine seconds.”
Cursor-Opus agent snuffs out startup’s production database

: Relax, the data's been recovered. Continue with your vibe coding

The Register