Agent incident database
Cursor agent deleted PocketOS's production database and backups in nine seconds
24 Apr 2026Anysphere (Cursor) · Cursor (running Claude Opus 4.6) with Railwaydestructive actionincident
An AI coding agent working on a staging problem at the software company PocketOS found an unrelated API token and used it to delete the company's production storage volume. Because the backups lived on the same volume, they were deleted too.
How it unfolded
Each step on its own looked like ordinary agent work. The harm came from the sequence.
- 1
Hit a credential mismatch while working on a routine task in the staging environment
- 2
Searched the codebase and found a Railway API token in an unrelated file
- 3
Ran a curl command against Railway's API using that token
- 4
Deleted the production volume, including the backups stored on it, with no confirmation step
harm
Which policy would have stopped this?
In our reading of the reports, these policies from the open library stop the harmful step before it runs.
- Destructive resource operations require a preceding gate: The agent's last step deleted a production volume. A destructive operation waits for a person's approval before it runs.
Open Destructive resource operations require a preceding gate in the playground and build the path yourself.
This is our reading of public reporting. We haven't run it against the original systems.
What happened
According to PocketOS founder Jeremy Crane, a Cursor agent running Anthropic's Claude Opus 4.6 hit a credential mismatch in the company's staging environment and decided on its own to fix it by deleting a Railway storage volume. It searched for an API token and found one in an unrelated file; the token had been created to manage custom domains through the Railway CLI but was scoped for any operation. The agent used it in a curl command that deleted the production volume without any confirmation step, and Crane said the deletion took nine seconds. Railway stored volume-level backups on the same volume, so they were erased as well, and Information Age reported the most recent recoverable backup was about three months old. Asked afterwards, the agent produced a written admission that it had ignored rules against destructive commands and had not verified that the action was scoped to staging.
Outcome
Months of reservations and customer sign-ups were initially lost, according to Crane. Railway CEO Jake Cooper stepped in two days later and helped restore the data within about an hour, according to The Register; Railway said it patched the legacy endpoint (which lacked delayed-delete logic) and added safeguards. The Register found no official Cursor response.
Sources
- Cursor-Opus agent snuffs out startup’s production database, The Register, 2026-04-27
- Gone in 9 seconds: AI agent deletes company database, Information Age (ACS), 2026-05-05
A summary of public reporting. The step-by-step timeline is our reading of those sources; where they differ, we say so above.
Your agents
Could this happen to yours?
For developers
Build a path yourself
Recreate a sequence in the playground and see how policies treat it.