Share
Anthropic’s Claude AI agent wiped the entire database and all backups of PocketOS, a SaaS (sofware-as-a-service) platform for car rental businesses, in a catastrophic event lasting just nine seconds.
The incident occurred when an AI agent running Claude Opus 4.6 via the coding tool Cursor was tasked with a routine maintenance job in a staging environment.
The disaster unfolded when the AI made a “guess” that deleting a staging resource would not impact production systems. Without verifying system architecture or asking for human confirmation, the agent issued a destructive command that targeted the live production database.
Because of the way the company’s cloud provider, Railway, handled storage, the single API call used by the AI to remove the main database simultaneously wiped all associated backups stored on the same volume.
Infrastructure failure
PocketOS founder Jer Crane noted that while the AI’s autonomous decision-making was the trigger, the lack of infrastructure safeguards turned a mistake into a total wipeout.
The system allowed high-impact deletions without confirmation and kept backups in the same environment as live data, leaving months of customer information unrecoverable in seconds.
The fallout has forced the startup into a gruelling manual recovery process. With the most recent viable backup being three months old, the team is currently attempting to reconstruct lost data using payment records, emails, and calendar integrations.
The event serves as a stark warning about the risks of “agentic” AI tools – systems granted the power to take high-impact actions without human checks or isolated “sandbox” environments.
Related Posts
Discover more from Tech Digest
Subscribe to get the latest posts sent to your email.