OpenAI's ChatGPT Desktop App Tracks User Activity

OpenAI has introduced a new feature called 'Computer History' for its ChatGPT desktop application on macOS, designed to record user interactions like clicks and keystrokes. This data is then used to train the AI models, learn user workflows, suggest automations, and potentially complete tasks left unfinished. The feature operates by building a timeline of user activity that can be referenced by ChatGPT and Codex when responding to prompts.

Ari Weinstein, Product and Engineering manager at OpenAI, stated on X that Computer History is designed to automatically disregard content viewed in incognito or private browser tabs, enhancing user privacy. Additionally, the feature is opt-in, meaning users must actively choose to enable it. For those who opt in, there are controls allowing them to exclude specific applications and websites from being tracked. Users also have the ability to delete individual entries or manage their data for more granular control over what information is collected.

Understanding the Data Collection

Dominik Kundel, a member of OpenAI's Developer Experiences team, demonstrated the feature's capabilities in a video. The demo showed the app recalling the last document edited by the user, checking if it had been shared via Slack, and providing a summary of the user's morning activities based on the recorded events. Unlike Microsoft's controversial Windows Recall feature, which relied on extensive screenshots, OpenAI emphasizes that Computer History does not capture images, videos, or audio. Instead, it focuses on 'events'—specific user actions and application interactions.

This development raises significant questions about data privacy and AI training. While OpenAI positions Computer History as a tool for personalized AI assistance and improved model training, concerns may arise regarding the extent of data collected and how it is secured. The company's emphasis on an opt-in model and exclusion options suggests an attempt to address these potential privacy issues proactively. However, the detailed nature of the data—tracking clicks and keystrokes—means users will need to carefully consider their privacy settings and the implications of enabling the feature.

The integration of Computer History into the ChatGPT desktop app marks a notable step in how AI tools interact with user environments. By analyzing activity patterns, OpenAI aims to make its AI more context-aware and helpful. The success and adoption of this feature will likely depend on user trust in OpenAI's privacy practices and the perceived value of the AI-driven assistance it provides. As AI continues to permeate personal computing, features like Computer History will become increasingly common, necessitating ongoing discussions about the balance between utility and user privacy.