OpenAI has rolled out a new feature in the ChatGPT desktop application for macOS that raises significant privacy concerns. The "Computer History" function allows the AI to track user actions, including keystrokes and clicks, to enhance its learning based on everyday computer interactions. According to OpenAI's product manager, Ari Weinstein, this capability enables ChatGPT to learn from everything users do on their devices. In a demonstration, developer Dominik Kundel showcased how the AI can access recently opened Google Docs and respond to inquiries regarding Slack messages by examining the user's activity history.
Importantly, the feature does not activate automatically; users must manually enable it. OpenAI claims to have implemented several protective measures: the function is designed to ignore private browsing modes, allows for the exclusion of specific applications from tracking, and enables users to view and delete saved "memories." However, vulnerabilities remain. The activity logs contain sensitive information that is not encrypted, making it potentially accessible to other applications on the same computer. Moreover, the AI can read messages in platforms like Slack without participants' explicit consent, leading to comparisons with Microsoft's similar Recall feature, which faced backlash for similar privacy risks.
Currently, the Computer History feature is exclusive to users with Pro, Business, or Enterprise subscriptions and is not available for free accounts. It is also inaccessible to users in the European Union, the United Kingdom, and Switzerland, likely due to stricter data protection regulations. Unlike Recall, which takes and stores screenshots on the device, OpenAI processes input events on its servers, prompting questions about data storage locations and durations.
Another concern highlighted by OpenAI is the risk of "prompt injection." If a user visits a site with hidden malicious instructions, ChatGPT or its Codex agent could inadvertently execute them, as everything within the function's scope becomes part of the model's context.
This new feature not only underscores the growing capabilities of AI technologies but also raises critical discussions about user privacy and data security. Competitors in the AI space will need to address these issues carefully to maintain user trust and comply with regulatory standards.
Informational material. 18+.
