ChatGPT's Secret Watch 👀: Privacy Nightmare? 😱

COMPUTER HISTORY FEATURE LAUNCHED: EXTENDING CHATGPT’S MEMORY CAPABILITIES
OpenAI’s ChatGPT and Codex have been leveraging “memories,” or contextual information derived from previous conversations, since April 2024. The newly launched Computer History feature expands this capability by allowing users to opt-in to track their interactions with websites and applications. This tracking aims to provide a deeper understanding of user intent, essentially following their digital footsteps to anticipate their needs. The feature initially launched this week on MacOS for subscribers to the Pro, Business, and Enterprise plans. Importantly, administrators of the Business and Enterprise plans have control over whether users can enable the feature, alongside individual user opt-in settings. This layered approach allows for granular control over data collection and usage.

KEY FUNCTIONALITY AND TECHNICAL DETAILS
The Computer History feature utilizes MacOS’ Accessibility Application Programming Interface (API) to capture a wide range of user activity, including clicks, typing, keyboard shortcuts, app switches, and contextual information. Unlike previous iterations like Windows Recall and OpenAI’s Chronicle preview, this implementation avoids screen captures, voice recording, and other intrusive monitoring methods. Instead, it focuses on the raw data exposed by the API. This approach allows the system to identify repeatable tasks based on historical activity, as demonstrated by the example of creating a video upload skill. The system synthesizes this activity into summaries, outlining what the user has done rather than what they saw or said. This means the system can track file locations without necessarily accessing the content of those files. Crucially, if the content is subsequently used in a chat interaction, the system can link the activity to that content. The feature operates by generating tokens for summarization and memory creation, acknowledging that the user’s plan’s token budget and context window size may influence its capabilities.

PRIVACY CONSIDERATIONS AND SECURITY IMPLICATIONS
OpenAI emphasizes that Computer History files can contain sensitive information and are not encrypted by the service itself. Other programs running as the user’s macOS account could potentially access these files, presenting a significant security consideration. While MacOS’ built-in disk encryption provides a layer of protection, the potential for misuse remains. OpenAI states it doesn’t retain the activity files unless legally mandated and doesn’t use them for training. Despite these assurances, the feature’s functionality raises concerns about privacy-slash-security complacency, mirroring anxieties surrounding browser fingerprinting and deanonymization. The summaries are generated in the cloud but stored locally on the user's system. The company explicitly warns users about these risks and highlights the potential for exploitation, even from seemingly innocuous data like keyboard events.