AI agent wipes company database in 9 seconds and then apologizes

An AI coding agent powered by Anthropic's Claude Opus accidentally deleted an entire company database - and all its backups - while attempting to fix a routine issue, without human approval.

The incident affected PocketOS, a software provider for car rental companies, causing a 30-hour outage and temporarily erasing customer records and bookings.

According to PocketOS founder Jer Crane, the AI ignored built-in safeguards that should have blocked destructive actions without user confirmation. When asked to explain its actions, the agent apologized, admitting it had violated its own safety rules.

"It took nine seconds," Crane wrote on X. "The AI decided on its own to delete the database instead of asking or finding a non-destructive solution."

Crane argued the incident highlights broader weaknesses in today's AI infrastructure, warning that companies are deploying autonomous AI agents faster than they are building adequate safety systems.

PocketOS later confirmed that all lost data was successfully restored.