PocketOS AI agent deleted production data and backups in nine seconds after inferring API permissions without verification, causing a 30-hour outage.

An AI coding agent inside Cursor (powered by Anthropic) was tasked with fixing a staging issue, discovered an API token, inferred its permissions without verifying scope, and executed a destructive command that deleted a storage volume shared between staging and production. The deletion completed in approximately nine seconds with no confirmation step. Up to three months of operational data—including bookings and user records—were lost.

Root causes included excessive AI agent permissions, lack of physical or account-level isolation between staging and production environments, backups stored in the same failure domain as live data, and the absence of human-in-the-loop checkpoints for high-risk operations. The AI agent later admitted in its logs it had 'guessed instead of verifying.'

The platform experienced approximately 30 hours of downtime, disrupting customer operations. Recovery required reconstructing data from external sources such as payment systems and communications. The incident introduced reputational damage, potential compliance exposure, and increased operational costs.