Kaspersky is highlighting potential privacy and security risks associated with Computer History, a feature in the ChatGPT app for macOS that allows the AI assistant to build a broader picture of users’ computer activity. As AI tools become increasingly integrated into everyday workflows, the feature raises important questions about how personal and professional information is collected, stored and protected.
Similar capabilities are emerging across the technology industry, including tools from Microsoft and Anthropic, reflecting a broader trend toward AI assistants that can better understand users’ activities and provide more contextual responses.
Introduced in mid-August, Computer History is available to ChatGPT Pro subscribers and higher-tier users. The feature is disabled by default and requires users to activate it and grant the necessary macOS permissions before it can operate.
Once enabled, the feature uses macOS accessibility capabilities to capture information such as text displayed on the screen, window titles, text entered into fields and interactions with applications. According to the feature’s stated limitations, it does not record audio, access masked password fields or view the contents of images and videos. OpenAI also says that private browsing windows are excluded.
Raw activity data remains on the Mac for up to 48 hours. During this period, ChatGPT periodically generates summaries of related activity, creating diary-like records of recent computer use. Kaspersky notes that these summaries are stored as plain-text files in a folder accessible to other applications on the device. When the summaries are subsequently incorporated into ChatGPT conversations, the platform’s standard privacy settings apply.
“The appeal is easy to understand because an AI assistant that knows what you have been working on can respond with much more useful context,” says Brandon Muller, Senior Security Consultant for the META region at Kaspersky. “The trade-off is that information from across your day is being brought together and summarised in one place. That can become valuable to somebody who should not have access to it.”
Local Data Storage Could Create New Security Risks
Kaspersky warns that the feature could create an additional target for macOS infostealers, a type of malware designed to search devices for valuable locally stored information. As new sources of data become available, attackers may adapt their techniques to exploit them.
Even without directly exposing passwords or payment card details, a summary of computer activity could reveal sensitive information about a user’s work, communications and recent actions. Such details could help cybercriminals develop convincing, highly targeted scams.
The concerns echo earlier debates surrounding Microsoft’s Recall feature, which is designed to make previous computer activity easier to retrieve but operates differently by capturing screenshots and analysing their contents. Microsoft introduced additional safeguards before Recall’s release in 2025, including encryption and user-controlled activation. Nevertheless, Kaspersky continued to advise caution for people who regularly work with sensitive information.
Kaspersky’s Recommendations for Mac Users
For users who choose to enable Computer History, Kaspersky recommends carefully controlling which applications and websites the feature can track. Platforms containing private conversations, financial details or confidential business information should be manually added to the exclusion list.
Users should also review their activity history regularly to identify any sensitive information that may have been captured unintentionally. Protecting the Mac against unauthorised access and malware with an appropriate security solution, such as Kaspersky Premium for macOS, is another recommended precaution.
For professionals who handle confidential client records, medical information or other protected data, Kaspersky advises significantly restricting the feature’s use or keeping it disabled altogether.
“Features like this show where AI assistants are heading. The more context they have, the more useful they can become, but that context also deserves protection,” Muller says. “Users should understand what is being collected, exclude information they would not want exposed, and make the choice based on the kind of work they do on that machine.”
