STUPID-2026-0057

Gemini CLI destroyed a user's project files after a failed mkdir, then confessed 'gross incompetence'

10.0critical
July 26, 2025Verified

  1. Instruction given

    Reorganize the files in a project directory.

  2. Expected behavior

    Verify each filesystem operation succeeded before acting on it; never move files into a directory that was never created.

  3. Actual behavior

    Gemini CLI misinterpreted a failed directory-creation command and never performed a read-after-write check to confirm the mkdir worked. Trusting its own actions implicitly, it moved files into a nonexistent location, irrevocably destroying them. It then confessed: 'I have failed you completely and catastrophically.'

  4. Damage

    A user's project files were irrecoverably lost because the agent acted on the assumption that its own failed command had succeeded — no verification, no undo.

In July 2025, product manager Anuraag Gupta documented how Google's Gemini CLI destroyed his project files while attempting to reorganize a directory. The agent issued a mkdir that failed, but never performed a read-after-write check to confirm the directory actually existed — it trusted its own action implicitly and moved files into a location that wasn't there, irrevocably destroying them. The tool then confessed in plain language: 'I have failed you completely and catastrophically,' admitting its own 'gross incompetence.' Coming days after Replit's production-database deletion, it reinforced a pattern across vendors: coding agents that assume their filesystem operations succeeded, with no verification step between a failed command and a destructive follow-up.

Classification

Agent
Gemini CLI
Failure mode
Destructive Action
Root cause
Tool Misuse
Domain
Infra
Source
News Report

Related incidents

10.0
Gemini CLI silently executed arbitrary code from an untrusted repo (CVE-2026-12537, CVSS 10.0)
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
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.