User Control

Privacy and Data Controls in ChatGPT Atlas

Balancing Intelligence with User Agency

An AI that understands everything you browse raises immediate privacy questions. OpenAI appears acutely aware of this tension. ChatGPT Atlas for macOS is built with privacy controls at its core, offering users granular authority over what ChatGPT can perceive, remember, and learn.

The philosophy seems to be: maximum capability, but only with explicit consent and ongoing user control. Every feature that increases ChatGPT's awareness comes with corresponding mechanisms to limit, review, or disable it.

Layers of Privacy Control

Page Visibility Toggle

Every site you visit in ChatGPT Atlas for macOS has a visibility setting accessible from the address bar. When visibility is enabled, ChatGPT can read page content and offer contextual assistance. When disabled, ChatGPT cannot see what you're viewing, and no browser memories are generated.

Browser Memories Management

Browser memories are entirely optional. When enabled, they allow ChatGPT to retain insights across sessions. Users can review all memories in settings, archive those no longer relevant, and delete browsing history to remove associated memories.

Incognito Mode

Opening an incognito window in ChatGPT Atlas temporarily logs you out of ChatGPT entirely. No memories are created, no context is shared, and no data is retained from incognito sessions.

Abstract visualization representing privacy controls and data protection in ChatGPT Atlas for macOS

Data Usage and Training in ChatGPT Atlas

By default, OpenAI does not use the content you browse in ChatGPT Atlas for macOS to train its models. This is an opt-in choice. If you choose to enable training on your browsing data, you can do so in data controls settings.

Important Note

Even if you opt into training, webpages that have opted out of GPTBot crawling will not be used. Business and Enterprise content is never used for training, regardless of user settings.

If you've previously enabled training for chats in your ChatGPT account, that setting carries over to ChatGPT Atlas conversations. This includes website content attached via the sidebar and browser memories that inform your chats.

Parental Controls in ChatGPT Atlas

Setting Inheritance

If a parent has configured parental controls for ChatGPT, those settings automatically apply to ChatGPT Atlas conversations.

Memory Restrictions

Parents can disable browser memories entirely for younger users, preventing any retention of browsing patterns.

Agent Mode Control

Parents can turn off agent mode, ensuring ChatGPT Atlas remains a passive assistant rather than an autonomous actor.

Site Permissions and System Access

Beyond AI-specific controls, ChatGPT Atlas for macOS follows standard browser permission patterns. Camera, microphone, and location access are requested per-site. Permissions are visible in the address bar when active, and can be modified or revoked at any time through site settings.

During onboarding, ChatGPT Atlas requests permission for notifications. All other system permissions are requested only when a site needs them. macOS-level permissions for camera or microphone can be managed through System Settings if needed.

"You're in control of what ChatGPT can see and remember as you browse." OpenAI, on the privacy philosophy of ChatGPT Atlas for macOS

Related topics about ChatGPT Atlas for macOS

What context awareness means → Agent mode safety measures → macOS system integration →