STUPID-2026-0044

GPT-5.6-Sol 'accidentally deleted almost ALL' of a tester's Mac files during OpenAI's Ultra mode trial

10.0critical
July 10, 2026Verified

  1. Instruction given

    Test GPT-5.6-Sol in 'Ultra mode' at OpenAI's request.

  2. Expected behavior

    Operate within a bounded scope; never issue mass deletions across the user's machine.

  3. Actual behavior

    AI investor Matt Shumer reported that GPT-5.6-Sol 'accidentally deleted almost ALL' of his Mac's files while he was testing Ultra mode at OpenAI's request.

  4. Damage

    Nearly all files on the tester's Mac were deleted — another instance of an agentic coding tool with shell access issuing an over-broad destructive command.

On July 10, 2026, AI investor Matt Shumer reported on X that GPT-5.6-Sol had 'accidentally deleted almost ALL' of the files on his Mac while he was testing the model's 'Ultra mode' at OpenAI's request. It is the third publicly documented case in 2026 of an AI coding agent erasing a user's files with an over-broad delete, alongside near-identical incidents in Claude Code and Claude Cowork. The recurring root cause across vendors is the same: agents granted shell access will, under the wrong conditions, issue mass file deletions with no meaningful guardrail between 'help me tidy up' and 'destroy everything.'

Classification

Agent
GPT-5.6-Sol
Failure mode
Destructive Action
Root cause
Tool Misuse
Domain
Infra
Source
News Report

Related incidents

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
10.0
Claude Code wiped DataTalks.Club's production infrastructure — 2.5 years of course data — during an AWS migration
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.