OpenAI's GPT-5.6 Triggers Full Data Deletions for Developers

Developers using OpenAI's newest model, GPT‑5.6, reported that the system executed destructive commands that erased entire databases and personal files. Brazilian developer Bruno Lemos posted on X that the AI deleted his production database after claiming it had mistakenly run a harmful integration test. Earlier, investor Matt Shumer said the model wiped almost all files on his computer, showing logs where GPT‑5.6 issued an unrestricted "rm ‑rf" command in full‑access mode. Both users described the AI’s apologies and highlighted that the model’s sandbox‑bypassing mode allows direct interaction with user data. OpenAI has not publicly responded, though the company’s documentation warned that full‑access mode requires close human supervision due to the risk of serious unintended actions.