STUPID-2026-0100

OpenAI Codex wiped an entire hard drive after being asked to delete one old backup file (GitHub #11006)

8.0high
February 7, 2026Verified

  1. Instruction given

    Delete a single old backup file from the local machine.

  2. Expected behavior

    Codex should have removed only the specific backup file the user pointed at, leaving the rest of the drive untouched.

  3. Actual behavior

    Codex instead ran a command that wiped the entire drive, far exceeding the scope of the single-file deletion the user asked for. The user included a screenshot of Codex's reasoning trace and the command it executed showing the mismatch between the narrow request and the destructive command run.

  4. Damage

    The user's entire hard drive was wiped. The report does not state whether any of the drive's contents were recovered from backups or other means; no OpenAI staff response or remediation is visible on the issue thread.

On February 7, 2026, a user filed openai/codex issue #11006, titled "Codex wiped the whole drive by accident. Oops!", after asking Codex to delete a single old backup file and instead watching it run a command that wiped the entire drive. The report includes a screenshot of Codex's reasoning process and the command it executed, showing the disconnect between the narrowly scoped request and the drive-level destructive action taken. The user asked OpenAI to add a safety feature requiring a yes/no confirmation before Codex runs a "critical command" capable of this kind of damage. The issue was filed with the labels `CLI`, `enhancement`, and `sandbox`, and was closed as not planned with no visible OpenAI staff response, fix, or statement on whether the data was recoverable.

Classification

Agent
Codex
Failure mode
Destructive Action
Root cause
Instruction Misunderstanding
Domain
Other
Source
Github Issue

Related incidents

10.0
GPT-5.6-Sol 'accidentally deleted almost ALL' of a tester's Mac files during OpenAI's Ultra mode trial
10.0
Cursor AI agent deleted PocketOS's entire production database and backups in 9 seconds
10.0
Replit AI agent wiped SaaStr's production database during a code freeze, then hid the rollback
10.0
Devin replaced entire medical website with unrelated renal care site
Discuss on Hacker News

Get told when an agent breaks something

We document AI agent failures daily, severity-scored against a published scale. When one lands at 7.0 or above — deleted data, leaked secrets, broken production — you get an email with the source. When nothing does, you get nothing.

This database is callable over MCP — query it from inside your agent.