OPENAI ADDRESSES GPT-5.6 FILE DELETION BUG

The company investigated multiple reports of unintended file deletions and found the problem concentrates in specific configurations. When full-access mode runs without sandboxing protections and without auto-review mechanisms, the risk increases significantly. OpenAI stated it is actively working to mitigate the issue. The findings suggest users can reduce risk by enabling sandboxing protections and auto-review features when running GPT-5.6 in full-access mode. This disclosure highlights the security considerations required when deploying advanced AI models with broad system access. The incident underscores the importance of safety constraints in AI systems, particularly when they have permission to modify or delete files on user systems. No timeline for a complete fix was provided.