The Dangers of AI Agents: Entire Firm’s Database Deleted by Claude-Based Agent
“Vibe coding,” or the flexibility to provide code utilizing a human language immediate, interpreted, and transformed into working code, has been the most important development of the previous yr. People are actually utilizing studying pc applications (known as brokers) to deal with duties kind of autonomously, whereas they deal with completely different elements of their enterprise or life.
But generally, issues go flawed. Horribly flawed.
AI Agent Goes Rogue: Kills Firm’s Database in Seconds
Nine seconds – that’s how a lot it took one AI agent to actually delete your complete codebase of an organization, leaving their customers with out entry to crucial knowledge.
PocketOS, a agency that gives software program for automotive rental companies, suffered a surprising outage over the weekend after an autonomous AI-based agent took inaccurate actions, wiping out their complete database and all backups in a complete of 9 seconds.
Yesterday afternoon, an AI coding agent — Cursor operating Anthropic’s flagship Claude Opus 4.6 — deleted our manufacturing database and all volume-level backups in a single API name to Railway, our infrastructure supplier.
According to the corporate’s founder, Jer Crane, they have been utilizing Cursor, powered by Claude Opus 4.6, extensively thought of probably the most distinguished mannequin for coding duties.
Crane defined that the agent was working “completely by itself initiative” and determined to repair an present drawback by simply deleting the database. They even requested this system to clarify why it did it, when it confessed in full and outlined all of the foundations that it broke.
Community Reacts
While the occasion is little question devastating for the individuals concerned, the neighborhood had blended reactions, with so much of the customers commenting beneath the thread stating that the primary mistake was trusting the AI agent with all these permissions on the similar time.
This isn’t only a “dangerous AI incident” , it’s a textbook enterprise failure throughout AI, safety, and infrastructure design. If something, the AI agent is simply the set off; the actual challenge is system design that allowed a single motion to wipe every little thing.
The put up The Dangers of AI Agents: Entire Firm’s Database Deleted by Claude-Based Agent appeared first on CryptoPotato.