OpenAI added a Computer History feature to the ChatGPT desktop app for macOS that records user activity in unencrypted local files [1, 2].
This development creates a potential security vulnerability because any software with access to the local file system could read the stored data. If sensitive information is captured during a session, it remains exposed on the hard drive without cryptographic protection [1, 3].
The feature is designed to allow ChatGPT to reference past user actions to provide more accurate and context-aware assistance [1, 3]. By tracking what a user does on their screen, the AI can better understand the workflow and provide tailored support based on previous activity [2].
However, the method of storage has drawn criticism from security researchers. The app tracks clicks and keystrokes, essentially acting as a local log of user behavior [2]. Because these files are not encrypted, the data is stored in plain text on the Apple computer [1].
Privacy experts said that this architecture deviates from standard security practices for handling sensitive user telemetry. While the data is stored locally rather than on OpenAI servers, the lack of encryption means the files are accessible to other programs running on the same machine [1, 3].
OpenAI has not yet provided a timeline for updating the encryption standards for these local logs. Users who are concerned about privacy may need to manage their app permissions or disable the history feature to prevent the recording of their desktop activity [1].
“OpenAI added a Computer History feature to the ChatGPT desktop app for macOS that records user activity in unencrypted local files.”
This incident highlights the tension between AI utility and system security. To make AI 'agents' more capable, developers are increasing the amount of telemetry the software collects from the OS. However, storing this high-fidelity data, which can include passwords or private messages, without encryption creates a new attack vector for malware on the user's device.
