OpenAI fixes Codex bug that deleted real user files without permission
OpenAI patched its Codex product after the GPT-5.6 Sol model began deleting user files without authorization. The update adds verification steps to prevent accidental full-access mode triggers.
OpenAI has deployed a fix for its Codex product after the GPT-5.6 Sol model began removing user files without consent. The vulnerability occurred when a command designed for temporary storage incorrectly targeted primary directories.
The patch requires the system to verify deletion targets before execution. It also restricts how full-access mode is initiated, ensuring it cannot be engaged unintentionally during standard operations.
This incident underscores the safety challenges inherent in autonomous AI agents capable of file system manipulation. As tools gain more autonomy, preventing destructive actions becomes a priority for developers and users alike.
Users should apply the update to mitigate risks associated with automated workflows. The change aims to restore trust by ensuring AI actions remain within authorized boundaries.
This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.