OpenAI has rolled out a new feature called "Computer History" on ChatGPT's desktop app for macOS that captures detailed information about user activity, including clicks, keystrokes, and document interactions, to help the AI assistant better understand how users work. The feature works on an opt-in basis, but its introduction has sparked privacy concerns among users who compare it to Microsoft's controversial Windows Recall feature.

What Exactly Is OpenAI's Computer History Feature?

The Computer History feature allows ChatGPT to learn from everything users do on their computers, according to Ari Weinstein, OpenAI's product and engineering manager. In a demonstration video, development experience engineer Dominik Kundel showed how the feature enables ChatGPT to answer questions about recently viewed documents and understand the context of user activities across multiple applications.

The feature works by giving ChatGPT access to user interactions on the device. When asked what document was last viewed, the AI can retrieve that information and even check whether the document was shared with others. The system builds a timeline of user actions that ostensibly helps the AI provide more contextual and helpful responses.

How Does Computer History Compare to Windows Recall?

Social media users have drawn comparisons between OpenAI's Computer History and Microsoft's Windows Recall feature, which faced significant backlash when it was discovered that the feature could inadvertently capture sensitive information like bank statements. However, there are important differences between the two approaches.

Windows Recall takes periodic screenshots of a user's desktop and uses artificial intelligence to scan text and create a searchable database of that visual content. OpenAI's Computer History, by contrast, does not capture screenshots. Instead, it focuses on recording events and interactions such as clicks, keystrokes, and other user actions. Despite this distinction, OpenAI itself warns that "Computer History files can contain sensitive information" and, critically, are not encrypted, meaning other programs running on the macOS user account may be able to access them.

What Privacy Controls Does OpenAI Offer?

OpenAI has built in several safeguards to address privacy concerns, though critics argue they may not be sufficient. Users have multiple options to manage their data collection:

• Opt-In Requirement: The feature is not enabled by default; users must actively turn it on to begin tracking their computer activity.
• App Blacklisting: Users can manually prevent specific applications from being tracked by Computer History, limiting what data the feature can access.
• Manual Data Review and Deletion: Users can view the data bank of information ChatGPT has collected about their activities and remove anything they do not want stored.
• Incognito and Private Browsing Exclusion: The feature will not access content viewed in incognito mode or private browser tabs, providing some protection for sensitive browsing.

"It lets ChatGPT learn from everything you do on your computer, so it can better understand how you work, finish tasks that you're in the middle of, and suggest skills and automations based on how you use your computer," said Ari Weinstein, OpenAI's product and engineering manager.

Ari Weinstein, Product and Engineering Manager at OpenAI

Why Are Privacy Advocates Concerned?

Despite OpenAI's assurances about privacy protections, several concerns remain unaddressed. The most significant issue involves third-party privacy: when ChatGPT accesses documents or messages shared between users, it may view information from people who have not consented to OpenAI's data collection. For example, if a user shares a private message or email with another person, ChatGPT could potentially access that shared content, but the other person involved has not agreed to have their information processed by OpenAI.

Additionally, the fact that Computer History files are not encrypted creates a vulnerability. OpenAI explicitly warns that "other programs running as your macOS user may be able to access them," meaning malware or other applications with system access could potentially read the stored activity data. Given the history of AI systems inadvertently accessing or misusing sensitive information, some users worry that such features may "accidentally-on-purpose" view details they are not supposed to access.

What Does This Mean for AI Model Training?

OpenAI has not clearly explained how data collected through Computer History will be used to train its foundational AI models. The company states that the feature helps ChatGPT understand how users work and suggest automations, but there is no transparent disclosure about whether aggregated, anonymized, or raw user activity data will be incorporated into future model training. This lack of transparency mirrors broader concerns about how AI companies use user data to improve their systems.

The introduction of Computer History reflects a broader trend in AI development where companies seek to gather more granular data about user behavior to improve model performance. However, this approach raises fundamental questions about the balance between AI capability and user privacy in an era when data breaches and unauthorized access to sensitive information remain persistent risks.