OpenAI's ChatGPT Desktop App Tracks Clicks and Keystrokes to Train AI

OpenAI has introduced a new feature called Computer History in its ChatGPT desktop application for macOS. The feature converts user activity into training data, learns how users work, suggests automations, and can even resume tasks left unfinished. This information was revealed by The Verge in a report published on August 17, 2026. Computer History uses user activity to build a timeline that ChatGPT and Codex can reference when users make requests. In other words, the system records a user's interaction history with their computer, including apps used, websites visited, and actions taken, then leverages that data to provide more contextual responses. The feature is designed as opt in, not opt out, meaning users must actively enable it.

OpenAI also provides granular controls, allowing users to exclude specific apps and websites from Computer History. Users can also delete individual entries for finer control over collected data. Ari Weinstein, Product and Engineering Manager at OpenAI, announced via his X account that Computer History will automatically ignore content in incognito or private browsing modes. This statement indicates OpenAI is attempting to balance functionality with user privacy considerations. The launch of Computer History marks a significant step in the evolution of AI assistants from mere responsive tools to proactive systems that understand user work patterns. By learning how users work, ChatGPT can suggest relevant automations and help complete unfinished tasks, creating a more integrated experience with daily workflows.

The feature also expands the capabilities of Codex, OpenAI's AI model designed for programming tasks. With access to user activity timelines, Codex can provide more contextual assistance in software development contexts, understanding ongoing projects and challenges users face. OpenAI's decision to make the feature opt in reflects awareness of user data sensitivity. In an increasingly strict regulatory landscape, especially in Europe with GDPR and other jurisdictions tightening privacy rules, this approach may help OpenAI avoid potential compliance issues. However, even though the feature is opt in, users who enable it should understand that their activity data will be used to train AI models.

OpenAI has not yet detailed how this data is processed, how long it is stored, or whether it can be permanently deleted from training systems. Computer History arrives amid intensifying competition in the AI assistant market. Other major tech companies are also developing similar features that leverage user context to improve AI response quality. OpenAI's emphasis on user control could become a competitive differentiator in an increasingly crowded market. From a technical perspective, the feature utilizes local and cloud processing capabilities to build activity timelines. Although full technical architecture details have not been disclosed, integration with Codex suggests OpenAI is building infrastructure capable of handling large scale contextual data.

For professional users such as software developers, data analysts, and content creators, this feature has the potential to significantly boost productivity. The ability to resume unfinished tasks and receive automation suggestions based on work patterns could reduce time spent on repetitive tasks. OpenAI has not announced when Computer History will be available on other platforms such as Windows or Linux. The Verge's report only mentions macOS availability, suggesting OpenAI may be testing the feature on a limited user base before broader rollout. The feature also raises data security questions. Although OpenAI states that incognito mode content will be ignored, users should still consider security risks if their devices are compromised.

Activity data collected by Computer History could be an attractive target for attackers seeking insights into user work habits and sensitive information. This development coincides with Google's move to allow users to remove visible watermarks from their AI generations, as reported by TechCrunch on August 14, 2026. While the two events are not directly related, they both indicate a broader industry trend toward greater user control over AI generated data and content. OpenAI's launch of Computer History demonstrates that the company sees strategic value in contextual data to improve AI product quality. With more data available for training, AI models can become more personalized and relevant to each user, creating a continuous improvement cycle.

There has been no official statement from OpenAI regarding monetization plans for this feature or whether collected data will be used for other commercial purposes. What is clear is that Computer History marks a paradigm shift in how AI assistants interact with users, moving from a reactive model to one that understands and anticipates needs.