OpenAI recently pushed an update to the ChatGPT application for the Apple macOS platform. However, one user reported that the updated AI assistant provided and indexed a complete list of local files and exact directory paths without authorization. This phenomenon quickly caused some users to panic about privacy security.
It is reported that when the user asked ChatGPT about the whereabouts of the missing items, AI accurately displayed the path to the local directory. From the perspective of security-conscious users, applications that access and index local files without explicit permission hit a security red line. This makes people wonder whether an application that has not obtained "full disk access" to the system has bypassed the platform's security protocols.
Although the accusation initially aroused some people's vigilance, most netizens on social platforms did not panic. Reddit community user "UnfoldedHeart" explained that in macOS and even Windows and Linux systems, unless the application is specifically restricted in a sandbox, it is completely normal system behavior to be able to obtain the regular directory path where the user's configuration file is located, because this is where various software, including ChatGPT, stores configuration information.
In addition, according to ChatGPT's risk statement, the application itself has the ability to read and edit files within its workspace and will request additional access permissions from the user when necessary. An AI agent cannot act against the user's wishes without obtaining permission. Another user named "fivetoedslothbear" also pointed out that this user may be using an advanced mode similar to "ChatGPT Codex", which is executing commands on the computer (similar to executing the "ls" command in the terminal) to understand what files are there, rather than illegally bypassing privacy settings in the background.
Industry opinion believes that taking into account Apple's consistently strict privacy control policies, the possibility of allowing third-party applications to easily bypass security mechanisms is extremely low. Judging from the current public feedback on social forums, more than half of the participants do not agree with this accusation of privacy leakage and believe that this is just the normal functional operation mechanism of the application.