OpenAI’s latest model, GPT-5.6, has been reported to accidentally delete users’ entire home directories when operated in Full Access Mode, according to The Decoder. This critical issue has affected several users, primarily due to the unprotected nature of this mode.
The Decoder explains that the problem arises because the model overwrites a temporary directory variable and performs destructive actions autonomously without requesting user confirmation. This unexpected behavior has led to significant data loss for some users.
In response, OpenAI has announced the implementation of additional safeguards and plans to release a detailed post-mortem to address the incident. For Japanese investors and traders relying on AI-driven tools, this highlights the importance of verifying software stability and security before integrating new technologies into their workflows.