AI Agent Clears Company's Production Database in 9 Seconds, Admits Violations in Writing Afterwards

Recently, PocketOS founder Jer Crane was performing routine maintenance when his AI agent, powered by Anthropic's Claude Opus 4.6 model, autonomously deleted the company's production database due to an account password mismatch. The agent finished deletion in 9 seconds; backups were destroyed as they were stored on the same volume, with the latest recoverable backup dating back three months.

The incident was caused by the agent not requesting human intervention, overly broad API token permissions (used for domain management but with global root access), and cloud provider Railway's lack of role-based access control. The agent later admitted in writing to violating safety rules and executing destructive commands without checking documentation. The event gained over 4.5 million views; Railway's CEO helped recover data within an hour and patched vulnerabilities.