Logo
FrontierNews.ai

ChatGPT's New Memory Problem: What Your AI Assistant Is Learning About You

ChatGPT and other AI assistants are moving beyond simple conversation to build detailed profiles of who you are, what you care about, and how you spend your time. OpenAI, Meta, and Google are all expanding memory features that let their AI tools learn from your past interactions, connected services, and even your computer activity. Unlike the familiar debate about whether AI companies use your conversations to train future models, these new memory systems operate separately, creating a more persistent digital picture of you.

How Is ChatGPT Building a Profile of You?

On August 13, OpenAI rolled out a feature called Computer History for ChatGPT's desktop app on macOS. The optional feature lets ChatGPT and Codex, OpenAI's coding tool, access selected activity from your apps and websites as context for conversations. Instead of taking screenshots, Computer History records interaction events, which OpenAI says is a more privacy-conscious approach. However, the feature is only available to Pro, Business, and Enterprise users, and it's turned off by default.

The mechanics are worth understanding. Temporary interaction-event files stay on your Mac for up to 48 hours while OpenAI processes them on its servers to generate memories. The company says it doesn't retain those temporary files after processing unless required by law, and doesn't use them for training. However, the generated memory files remain on your Mac as plain-text Markdown files, unencrypted by default. This means other programs running under your macOS user account could potentially access them.

This is just one piece of a larger trend. Meta announced in 2025 that interactions with its generative AI features could become signals for the content and ads you see across Meta products. Those changes took effect on December 16, 2025, in most supported regions. For example, 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, including saved media like images, files, and audio or video recordings you upload while using supported Search services.

What Privacy Controls Actually Exist?

The good news is that you have options, though they vary by service. OpenAI has separate controls for model training, memory, and advertising personalization, meaning turning off model training doesn't necessarily turn off memory or personalization. With Computer History, you can decide which apps and websites contribute information, pause collection, inspect your timeline, and delete history.

Meta takes a different approach for users who want more privacy. On May 13, 2025, Meta announced Incognito Chat for Meta AI on WhatsApp and in the Meta AI app. The company says Incognito Chat processes conversations inside a secure environment that Meta itself cannot access, conversations are not saved, and messages disappear by default. Meta also places limits around sensitive subjects, saying it does not use information from AI conversations about health or political views to show you ads, and excludes religious views and other sensitive categories from ad personalization.

Google's approach requires more manual management. If you turn off the Save Media setting in Search Services History, Google stops saving media from future Search interactions. However, previously saved media doesn't disappear when you simply turn the setting off. You need to delete it separately if you want it removed from your account. Media already selected for AI training may remain in Google's training systems after it has been disconnected from your account.

How to Manage Your AI Privacy Settings

  • Review Your Service Settings: Check the privacy and ad settings for ChatGPT, Meta AI, and Google Search before sharing personal information. Each service has different controls for memory, training, and personalization.
  • Disable Features You Don't Need: Computer History is off by default for ChatGPT, and you can choose which apps and websites contribute information. Turn off Save Media in Google Search Services History if you don't want media saved from your interactions.
  • Use Private Alternatives When Available: Meta's Incognito Chat offers a more private conversation experience where messages don't persist and Meta cannot access the conversation. Consider using this feature for especially sensitive topics.
  • Be Selective About What You Share: The more information an AI service can connect about you, the more personalized it becomes, but also the more detailed your digital profile becomes. Avoid sharing sensitive health, financial, or personal information unless necessary.
  • Delete Previously Collected Data: Simply changing a privacy setting and deleting previously collected data are not the same thing. Actively delete data you no longer want stored, as it may not disappear automatically.

Why This Matters Beyond Training

For years, the biggest AI privacy concern has been whether companies use your conversations to train future models. That still matters, but it's only part of the picture. AI assistants now use memory to personalize future conversations, draw context from connected services, and use certain AI interactions to help decide what content or ads you see. This creates a feedback loop where your AI activity influences what you see online, which in turn influences what the AI learns about you.

The shift reflects how AI assistants have become more capable and integrated into daily life. ChatGPT Work, for example, can browse the web, manage files, and complete complex tasks over multiple steps, much like an AI agent would. This expanded functionality means these tools have access to more of your digital life than ever before. A user researching wireless mice on Amazon, for instance, might ask ChatGPT Work to find specific products, open them in tabs, and check shipping availability, all while the AI learns about their shopping preferences.

The practical implication is clear: as AI assistants become more capable and integrated, being selective about what you share becomes even more important. The more information an AI service can connect about you, the more personalized it can become, which can be useful. But it also creates a much more detailed digital picture of your interests, routines, and concerns.