OpenAI responds to GPT-5

OpenAI responded to the GPT-5.6 Codex deletion controversy: it was an "accidental error" in the AI ​​operating environment settings. OpenAI officially confirmed that under certain conditions of use, the Codex code generation system driven by the GPT-5.6 model had a problem of accidentally deleting user local files. OpenAI core product lead Tibo Sottiaux publicly responded to the situation on social media platform According to preliminary analysis, this data loss risk mainly occurs when the user turns on the "full access mode" and does not enable the sandbox protection mechanism. In these specific settings, when the AI ​​model performed the task, it was intended to try to define a temporary working directory by overwriting the environment variable $HOME. However, an error occurred during the actual operation, causing the system to delete the entire $HOME directory instead. For macOS and Linux users, this means that the user’s home directory and the important personal data within it are at risk of being wiped out. In addition, Sottiaux specifically pointed out that the problem is more likely to be triggered when the "automatic review function" is disabled. Sottiaux attributed the incident to an "accidental error" in the AI ​​system. In response to this risk, OpenAI is currently taking a number of emergency mitigation measures. The company is updating the developer prompt information in "Full Access Mode" to clearly inform users of the operational impacts they may face when granting this permission, and to guide users to switch to a more secure permission mode. At the same time, OpenAI has also reinforced additional protective measures in the system to prevent such misoperations from happening again. While OpenAI emphasized that cases affected by this vulnerability are extremely rare, the company promised to release a detailed post-mortem report (post-mortem) in the next few days to disclose more technical details and explain how to further reduce such risks. For users who are currently using Codex and have been granted system-level access, OpenAI recommends immediately revoking this high-privilege mode and switching to a restricted operating mode that includes necessary security protections. At the same time, the new hardware product that has attracted much attention recently, the Codex Micro physical control panel priced at US$230, is currently sold out according to the official page. Although this product is mainly aimed at developers, its design language and interaction logic are regarded by the outside world as a reference prototype for future OpenAI consumer hardware product design. via cnBeta.COM - Chinese Industry Information Station (