OpenAI has quietly expanded ChatGPT's ability to remember and learn from your personal activity, introducing a new Computer History feature that tracks what you do on your Mac. The optional feature, which rolled out on August 13, lets ChatGPT and Codex access selected activity from your apps and websites as context for conversations. While the capability offers genuine convenience, it also highlights a growing privacy tension: as AI assistants become more personalized, the digital picture they build of you becomes far more detailed.

What Exactly Is ChatGPT's Computer History Feature?

Computer History is a new addition to the ChatGPT desktop app for macOS that works by recording interaction events from your selected apps and websites, rather than taking screenshots or screen recordings. OpenAI says the feature excludes private browsing and gives you granular control over which apps and websites contribute information. You can pause collection, inspect your timeline, and delete history at any time.

However, access is currently limited. OpenAI offers Computer History only to Pro, Business, and Enterprise users. Business and Enterprise administrators must enable access before individual team members can turn it on. The feature is off by default, and it requires the Memories feature to function.

The technical mechanics reveal important privacy details. OpenAI says temporary interaction-event files remain on your Mac for up to 48 hours while the company processes them on its servers to generate memories. The company states it does not retain those temporary files after processing unless required by law, and does not use them for training. However, the generated memory files stay on your Mac as plain-text Markdown files until you delete them. Critically, Computer History does not encrypt these files itself, meaning other programs running under your macOS user account may be able to access them.

Why Are AI Companies Expanding Memory and Personalization Features?

The shift toward deeper personalization reflects a broader industry trend. For years, the main privacy concern around AI assistants centered on whether companies use your conversations to train future models. That question still matters, but it is only part of the picture. Today, AI assistants use memory to personalize future conversations, draw context from connected services, and use certain interactions to influence what content or ads you see.

This means turning off model training does not necessarily disable memory or every form of personalization. OpenAI, for example, has separate controls for model training, memory, and advertising personalization. Meta takes a more direct approach, announcing in 2025 that interactions with its generative AI features could become signals for the content and ads people see across Meta products. Those changes took effect on December 16, 2025, in most supported regions. For instance, chatting with Meta AI about hiking could signal interest in hiking, leading to related recommendations or advertisements.

Google is also expanding what can become part of your Search Services History. If Search Services History and its Save Media setting are enabled, Google can save media from your Search interactions, including images, files, and audio or video recordings you upload or create while using supported Search services. Google says saved media can help personalize your experience and may be used to develop and improve its AI models. If Google selects saved media for AI training, it disconnects that media from your Google Account first, though training data may remain for up to four years.

How to Manage Your Privacy Across AI Assistants

• Review Separate Controls: OpenAI, Meta, and Google each have distinct settings for model training, memory, personalization, and advertising. Disabling one does not automatically disable the others, so check each service individually to understand what data is being collected and how it is being used.
• Limit App and Website Access: With ChatGPT's Computer History, you decide which apps and websites contribute information. Exclude sensitive applications and websites from the feature, and regularly inspect your timeline to see what has been recorded.
• Delete Previously Collected Data: Simply turning off a privacy setting does not remove data already collected. You must manually delete previously saved media, memories, and history if you want them removed from your account.
• Use Private or Incognito Modes When Available: Meta offers Incognito Chat for Meta AI on WhatsApp and in the Meta AI app, a feature that processes conversations in a secure environment that Meta itself cannot access, with messages disappearing by default.
• Be Selective About What You Share: As AI assistants learn more about you, being thoughtful about what personal information you share becomes increasingly important, especially regarding health concerns, financial decisions, or other sensitive topics.

Meta places limits around certain sensitive subjects. The company says it does not use information from AI conversations about health or political views to show you ads, and it excludes religious views and several other sensitive categories from ad personalization. Still, regular Meta AI interactions can become another source of information used to personalize your experience.

What About ChatGPT Work and Agentic AI?

Beyond memory and personalization, OpenAI has also expanded ChatGPT's capabilities through ChatGPT Work, an agentic AI tool that can browse the web, manage files, and complete multi-step tasks autonomously. ChatGPT Work is available to all ChatGPT users, including those on the free tier, though free users have limited capacity. Paid plans offer significantly more usage capacity.

Agentic tools like ChatGPT Work consume substantial computational resources, measured in tokens. A user on the $20 per month ChatGPT Plus plan can typically complete a fairly large project every few days without hitting usage limits, with roughly 20 hours of agentic work available per month. Higher-tier plans provide more capacity.

For users considering ChatGPT Work, practical caution is essential. One expert recommends feeding the AI small assignments one at a time, asking for a plan before execution, and checking usage availability before starting large projects. Never provide passwords to the AI, and always maintain backups before allowing it to reorganize or delete files. Most importantly, double-check everything the AI produces rather than assuming accuracy.

The expansion of ChatGPT's capabilities reflects a fundamental shift in how AI assistants operate. They are no longer simple question-answering tools; they are becoming persistent agents that learn about you over time, access your activity, and use that knowledge to shape your experience. As these tools become more capable and more integrated into daily life, understanding and managing your privacy settings is no longer optional but essential.