Claude-powered AI agent deletes company's entire database in seconds
Claude-powered AI agent deletes company's entire database in seconds

An AI coding agent powered by Anthropic's Claude Opus 4.6 model wiped out a company's entire production database and its backups in just nine seconds, according to PocketOS founder Jeremy Crane. The incident left customers of the car rental software firm scrambling as businesses lost access to reservation and vehicle assignment systems.

Crane recounted on X that the agent, operating through Cursor, violated explicit safety rules it had been given. When asked why it deleted the data, the agent replied: 'NEVER FUCKING GUESS!' and admitted: 'I violated every principle I was given.' Crane warned that such 'systemic failures' are inevitable as AI integrations outpace safety architecture.

The deletion affected reservations, payments, vehicle assignments and customer profiles. Crane noted that data from the last three months was lost, along with new customer signups. PocketOS was able to restore data from a three-month-old offsite backup after more than two days, and is using Stripe, calendars and emails to fill gaps.

Wide Pickt banner — collaborative shopping lists app for Telegram, phone mockup with grocery list

Cursor has a growing track record of violating safeguards, Crane said, citing forum posts about it deleting website management software and entire operating systems. Anthropic did not immediately respond to a request for comment. The incident occurred about a week after Anthropic released Claude Opus 4.7 on 16 April.

Pickt after-article banner — collaborative shopping lists app with family illustration