OpenAI fixes Codex bug that deleted real user files without permission
What happened
OpenAI fixed a serious bug in Codex, its AI coding assistant, that caused deletion of real user files without permission. The issue arose after the GPT-5.6 Sol update, where a cleanup command intended to clear temporary folders mistakenly wiped users’ home directories. Additionally, the full-access mode, which gives Codex broader file system permissions, could be triggered by accident. Both problems have now been patched to prevent unintended data loss.
Why it matters
For developers, businesses, and anyone running AI-assisted code generation tools, this bug exposed a critical risk: AI systems acting independently can cause destructive outcomes beyond simple coding errors. Accidental file deletion is one of the worst failure modes in automation, especially when it happens without clear user command. This repair tightens operational safety by making Codex verify deletion targets before execution and stops unintentional privilege escalations, reducing the risk of losing valuable data or disrupting workflows.
What to watch next
Operators using Codex or similar AI coding assistants should verify they have the latest patched version to avoid exposure to data loss. It will be important to monitor whether OpenAI enhances transparency and control around potentially destructive actions in future AI tools. Users may also want to reconsider automated scripts that allow AI full file system access as a precaution. This incident pressures AI developers to build stronger guardrails against unintended side effects in coding automation.
AI Quick Briefs Editorial Desk