OpenAI upgrades Codex security protection, adding dangerous operation interception and permission control OpenAI announces the upgrade of the security protection system of its AI programming agent Codex, using multiple security measures to reduce the risk of potentially destructive operations such as accidentally deleting files and overwriting data when the agent executes code tasks.

OpenAI upgrades Codex security protection, adding dangerous operation interception and permission control OpenAI announced that it has upgraded the security protection system of its AI programming agent Codex, through multiple security measures to reduce the risk of potentially destructive operations such as accidental deletion of files and overwriting of data by the agent during the execution of code tasks. Tibo, a member of the OpenAI team, recently posted on social platforms that the company investigated the recent Codex destructive operation reports and found that the main risk is concentrated in the temporary working directory cleaning process. Part of the problem stems from the reuse of system variables. For example, incorrect use of variables such as $HOME causes the cleanup operation to point to the real user directory; in addition, some operations lack sufficient path verification before deletion or overwriting, which also increases the risk of data corruption. In response to the above problems, OpenAI has strengthened protection from multiple levels. The new version of Codex requires agents to verify the target path before performing sensitive operations and use a separate temporary directory to avoid relying on potentially misleading system variables. At the same time, OpenAI has strengthened the dangerous command identification and interception mechanism, and added relevant operation review processes. In addition, OpenAI also raises the threshold for using Codex Full access permissions, reduces the possibility of misauthorization by users by adding risk prompts, and upgrades Auto-review rules to further control security risks while maintaining AI programming efficiency. As AI programming agents gradually participate in code development, file management, and automated task execution, agent behavior security has become a focus of the industry. OpenAI This upgrade reflects the need for AI Agent to simultaneously improve its permission management, operation auditing and risk prevention and control capabilities in the process of expanding its application scope. via AI News (author: AI Base)