By skyforbes 
OpenAI’s launch of ChatGPT Atlas, a new AI-powered web browser built around the ChatGPT experience, represents what the company claims is the “next big leap” in browser innovation. Atlas fundamentally changes how users interact with the internet by incorporating features like a contextual sidebar companion, search powered by AI, and the capacity for ChatGPT to take actions autonomously via Agent Mode.
However, one of the most unique and potentially controversial features is its highly personalized browser memory. While designed to make the browser more helpful, its comprehensive data collection has raised immediate questions about privacy, and what exactly it means for your browser to remember everything you do.
The power, and therefore, possibly peril, of Atlas’s memory
The memory feature is considered one of the most unique and powerful elements of the new browser. OpenAI developed Atlas with memory as a critical core feature, noting that as users utilize ChatGPT more, it becomes more personalized and useful, and this principle is now extended across the entire web browsing experience.
Here’s what Atlas remembers:
- Browsing history: Atlas remembers where you browsed on the web and what you were doing. It can even import your existing browsing history and bookmarks from Chrome during setup.
- Chat context: It remembers your past conversations in chat.
- Search and recall: The memory enables the AI to answer complex, context-dependent questions, such as helping you find an article you read days ago but cannot locate. For example, a user could prompt, “Yesterday I was looking at some knowledge management tools but I can’t remember what they were please help,” and Atlas will search its “browser memories” to pull up the exact pages viewed.
Why the memory feature is a “Red Flag”
While the utility of memory is clear, the fact that the browser is designed to remember nearly everything is a source of concern for many users. The source material notes that if you do not want Atlas to remember what you looked at, you must use incognito mode.
This extensive tracking also ties into how OpenAI uses user data. If a specific setting is enabled, your activity in Atlas could be used to improve the underlying models. This is particularly concerning for individuals who don’t want the information related to their browser activity to be “used for AI or to be sent to chat GPT for training”.
In addition to browser and chat history, users also face a decision regarding data sharing. Under Data controls, there is a setting labeled “Improve the model for everyone”.
If this feature is turned on, the browser states that it allows your content to be used to train OpenAI’s models, making ChatGPT better for all users.
For users sensitive to privacy, the automatic retention of this data constitutes a major “red flag”.
How to turn off the creepy memory features
Fortunately, OpenAI recognized the need for user control, and all these memory and data-sharing features are optional and can be disabled. If you decide that this level of permanent browser memory isn’t for you, you can manually turn off several key personalization settings.
To turn off the memory features in ChatGPT Atlas:
1. Navigate to your profile within the browser.
2. Go to Settings.
3. Click on Personalization.
4. Within the Personalization menu, you will find, and can then disable the following options:
o Reference saved memories.
o Reference chat history.
o Reference browser memories.
If you wish to prevent your data from being sent back to OpenAI for model training, an additional step is required under data controls:
1. Click on Data controls.
2. Turn off the setting labeled Improve the model for everyone.
Disabling these features means you can use the browser without your information being saved and referenced later.
Furthermore, if you only need temporary privacy for a specific session, you can always open an incognito window, which will ensure your activities are not remembered by ChatGPT.