OpenAI's ChatGPT Introduces 'Computer History' Feature, Shifting to Keylogging for User Data
OpenAI's new 'Computer History' feature for ChatGPT on macOS acts as an opt-in keylogger, capturing user input events and storing them locally before sending summaries to OpenAI.
OpenAI has introduced a new feature for its ChatGPT desktop application on macOS called 'Computer History.' This opt-in functionality shifts from its previous screenshot-based approach to a keylogging and event capture system. The feature aims to enhance ChatGPT's responses by creating a timeline of user interactions across various applications and websites, potentially surfacing automation opportunities and aiding in resuming prior work.
Unlike its predecessor, Chronicle, which relied on screenshots, Computer History focuses on recording input events such as clicks, typing, keyboard shortcuts, and app switches. This data is stored locally and unencrypted for up to 48 hours before being processed. OpenAI states that the feature does not capture screen images, microphone input, or system audio, nor does it record private-mode browsing. However, the company's own documentation warns that these local files "can contain sensitive information" and "other programs running as your macOS user may be able to access them."
The feature is available to ChatGPT Pro, Business, and Enterprise users on macOS. Pro users can enable it individually, while Business and Enterprise administrators must approve its use. It is currently not available in the European Economic Area (EEA), Switzerland, or the United Kingdom. Users accessing ChatGPT via API key or Amazon Bedrock are also excluded.
OpenAI provides specific guidance for users, advising them to suspend the service during communications with others unless explicit consent is obtained. Users are also encouraged to consider pausing the feature or excluding applications that handle sensitive health, financial, or personal information, acknowledging the potential privacy implications.
While OpenAI claims that event files are not retained after processing unless legally required and are not used for training, the company has historically provided chat logs in response to legal processes. The 'Computer History' feature also introduces additional costs, as it consumes tokens for summarizing activities and generating memory data. Furthermore, it expands the attack surface for prompt injection, as malicious instructions embedded in websites or applications could be followed by ChatGPT or Codex.
This shift towards capturing user input events represents a significant change in how OpenAI is seeking to gather contextual data for its AI models. The move from passive screenshotting to active input monitoring raises further privacy concerns among users and cybersecurity professionals, despite the opt-in nature and stated limitations.
The feature's expansion of the prompt injection risk is a notable cybersecurity implication. If a user visits a compromised website or opens a malicious document, the captured input events could inadvertently lead ChatGPT or Codex to execute harmful instructions, potentially compromising the user's system or data.
Ultimately, 'Computer History' positions user activity as a direct input for AI enhancement, blurring the lines between user assistance and pervasive surveillance. The unencrypted local storage and the potential for misuse, even with opt-in controls, highlight ongoing debates about data privacy and security in the age of increasingly integrated AI tools.