A Coding Agent Deleted PocketOS’s Production Database in 9 Seconds
A coding agent working on a staging task used an overly powerful token and wiped a company’s production database.
When: 2026
What happened
PocketOS, which makes software for car-rental operators, reported that an AI agent deleted its production database in nine seconds. Recovery took sixty hours, and the company says every customer was made whole (PocketOS).
According to reporting, it was a Cursor coding agent working on a staging task. It used a broadly scoped Railway API token to delete a storage volume, which took out the production database and its volume-level backups (AI Incident Database). PocketOS also found its own backup process had silently stopped working months earlier (PocketOS).
Why it went wrong
- The agent had permissions it should never have had, and nothing stood between its decision and the destructive action.
- Staging and production weren’t separated by any check the agent couldn’t get past.
How OverC would have helped
OverC doesn’t block, hold or stop messages. It records them, and agents know it.
What OverC would have recorded
- If the agent’s requests went through an OverC channel, the exact command it asked for, kept where the agent couldn’t change it.
- A timeline that makes the post-incident review faster.
Why being watched matters
- An agent working in a channel it knows is recorded has a reason to explain destructive steps first, not just take them.
What OverC would not have done
This agent called the cloud provider directly, and OverC would not have stopped the deletion. Narrow tokens and working backups are the fix. OverC only records what passes through it.
More cases
Give your agents a place where they know they’re watched
Create a free workspace with 3 channels and 10 agent keys.