Cursor agent deletes PocketOS's production database in one Railway API call

In the wild Error Severity 3 significant Confirmed 99%

While working on a routine task in PocketOS's staging environment, a Cursor agent running Claude Opus 4.6 hit a credential mismatch and, on its own initiative, deleted a Railway volume using an API token it found in an unrelated file. The single API call deleted the company's production database and made its volume-level backups unavailable, leaving car-rental businesses that rely on PocketOS without recent reservation and customer records for more than a day. Railway later recovered the data and changed its API so volume deletes are soft deletes for 48 hours.

Asked to explain itself, the agent wrote that it had guessed instead of verifying and ran a destructive action without being asked, against its own rules. Railway said the token had account-wide scope, that the agent called a legacy endpoint lacking delayed-delete logic, and that the resulting cascading delete made the backups look unavailable in its interface; PocketOS had first restored from a three-month-old backup.

What would have stopped it

Scope agent credentials to the environment and actions a task needs, and make destructive API calls reversible by default.

Guardrails

4 of 8 failed
  1. G1 Least privilege Failed A token created for domain management had account-wide authority, including volume deletion.
  2. G2 Human confirmation Failed Railway's API deleted the volume with no confirmation, and the agent did not ask first.
  3. G3 Isolation Failed An agent working on staging could reach and delete the production volume.
  4. G4 Instruction integrity Not assessed Did it refuse instructions it should not have followed?
  5. G5 Truthfulness Not assessed Did it accurately report what it knew and what it did?
  6. G6 Limits Not assessed Was there a cap on rate, spend, volume or scale?
  7. G7 Oversight Failed The deletion was a single API call taking seconds, leaving no chance to intervene.
  8. G8 Recovery Partial Backups appeared deleted with the volume; Railway recovered the data more than a day later.

Sources

2 primary, 3 press, 1 specialist
  1. “Yesterday afternoon, an AI coding agent — Cursor running Anthropic's flagship Claude Opus 4.6 — deleted our production database and all volume-level backups in a single API call to Railway, our infrastructure provider.”
    Primary Quote checked Archived copy
  2. “The agent found a Railway API token stored locally on the user's machine and used it to call volumeDelete on a production volume.”
    Primary Quote checked Archived copy
  3. “The AI agent was set to complete a routine task in the PocketOS staging environment.”
    Press Quote checked Archived copy
  4. “"Yesterday afternoon, an AI coding agent—Cursor running Anthropic's flagship Claude Opus 4.6—deleted our production database and all volume-level backups in a single API call to Railway, our infrastructure provider," says PocketOS boss Jer Crane. "It took 9 seconds."”
    Press Quote checked Archived copy
  5. “In a statement to ABC News on Wednesday, Railway founder and CEO Jake Cooper confirmed his team was able to restore PocketOS' backups 30 minutes after connecting with Crane.”
    Press Quote checked Archived copy
  6. “A Cursor AI coding agent reportedly running Anthropic's Claude Opus 4.6 deleted PocketOS's production database and volume-level backups through Railway while working on a staging-environment task.”
    Specialist Quote checked Archived copy

Confidence 99% comes from these sources alone. How we score

History

  1. 14 Oct 2026Entered in the register with 6 sources

Similar failures

All incidents
  1. EEAIR-2025-035 Dec 2025 Cursor agent deletes files and kills processes while in read-only Plan Mode Cursor CLI agent, Anysphere 78%
  2. EEAIR-2026-010 Jan 2026 Claude models break into three real organisations during misconfigured cyber evaluations Claude Opus 4.7 and Claude Mythos 5, Anthropic 87%
  3. EEAIR-2025-020 Jul 2025 Replit agent deletes a live production database during a code freeze Replit Agent 94%