An artificial intelligence coding assistant wiped out an entire company database — and all its backups — in just nine seconds.
According to The Independent, PocketOS, a software provider for car rental businesses, suffered the digital equivalent of pulling the fire alarm and flooding the building after an AI agent powered by Anthropic’s Claude model decided, on its own, that deleting the production database was the best way to solve a routine issue.
PocketOS founder Jer Crane reportedly said the company was using Cursor, a coding agent running Claude Opus 4.6, when the AI went rogue over the weekend. Instead of asking for confirmation before performing what most humans would classify as “the worst possible option,” the system simply hit delete.
“It took nine seconds,” Mr. Crane wrote in a post on X, noting that the AI later produced what can only be described as a guilty plea.
In its own written confession, the AI admitted it violated explicit instructions never to run “destructive/irreversible” commands without user approval.
“Deleting a database volume is the most destructive, irreversible action possible,” the bot reportedly wrote, adding, “I guessed instead of verifying.”
As a result, customer reservations from the past three months disappeared, along with new customer signups, leaving rental businesses scrambling… and one robot presumably updating its résumé.
Fortunately, by Monday, the company had recovered the data via an old-school backup system.