Privacy & Safety in ChatGPT Atlas

Understanding how your data is protected and controlled

Privacy-First Design

ChatGPT Atlas is built with user privacy and control as fundamental principles. Unlike traditional browsers where AI features might collect data without clear visibility, Atlas gives you complete transparency and control over what ChatGPT can see, remember, and use.

Core Principle: You decide what ChatGPT has access to, and you can change these decisions at any time.

Key Privacy Features

1. Browser Memories Control

Browser Memories is an optional feature that allows ChatGPT to remember context from websites you visit. This is completely separate from traditional browsing history.

What Are Browser Memories?

When enabled, ChatGPT creates memories of key information from your browsing sessions, such as:

  • Job postings you've reviewed
  • Products you've researched
  • Articles you've read on specific topics
  • Patterns in your web activity

How You Control Them:

  • Opt-In Only: Browser Memories are off by default. You must explicitly enable them.
  • Toggle Per Site: Use the address bar toggle to control which sites ChatGPT can see
  • View All Memories: Access a complete list of all browser memories in Settings
  • Archive Individual Memories: Remove specific memories that are no longer relevant
  • Clear All: Delete all browser memories at once if desired
  • Auto-Delete: Clearing browsing history automatically deletes associated browser memories

2. Page Visibility Controls

You have granular control over which pages ChatGPT can see and interact with.

The Visibility Toggle

Located in the address bar, this toggle lets you control ChatGPT's access to the current page:

  • Visibility ON: ChatGPT can see page content and create memories (if enabled)
  • Visibility OFF: ChatGPT cannot view the page content, and no memories are created

Use Cases for Visibility OFF:

  • Banking and financial websites
  • Medical or health information
  • Personal emails or messages
  • Any sensitive or private content

3. Incognito Mode

Atlas includes a full incognito/private browsing mode with enhanced privacy features:

What Happens in Incognito:

  • ChatGPT is temporarily logged out
  • No browsing history is saved
  • No browser memories are created
  • No cookies or site data persist after closing
  • No data is sent to OpenAI

When to Use Incognito:

  • Browsing sensitive topics
  • Shopping for surprises or gifts
  • Using shared computers
  • Any time you want complete privacy

4. Data Training Controls

OpenAI has clear policies about how your browsing data may or may not be used for training:

Default Behavior:

By default, content you browse is NOT used to train AI models.

Opt-In Training:

You can choose to opt-in to allow your browsing data to help improve ChatGPT:

  • Navigate to Data Controls settings
  • Enable "Include web browsing" option
  • This setting is synced with your ChatGPT chat training preferences

Important Protections:

  • Even if you opt-in, websites that have opted out of GPTBot will not be trained on
  • Browser memories that inform your chats may be included if chat training is enabled
  • You can change this setting at any time

Parental Controls

Atlas includes comprehensive parental controls that extend beyond standard ChatGPT protections:

ChatGPT Controls

Existing ChatGPT parental controls automatically carry over to Atlas conversations

Browser Memories

Parents can disable Browser Memories entirely for child accounts

Agent Mode

Parents can turn off Agent Mode to prevent autonomous web actions

Content Filtering

All ChatGPT safety features apply to web content interactions

Security Features

Standard Browser Security

Atlas implements industry-standard security practices:

  • HTTPS Enforcement: Automatic upgrades to secure connections when available
  • Certificate Validation: Warns about invalid or suspicious security certificates
  • Phishing Protection: Identifies and warns about known malicious sites
  • Secure Password Storage: Encrypted password manager with strong encryption
  • Automatic Updates: Regular security patches and improvements

AI-Specific Security

Additional protections related to ChatGPT integration:

Prompt Injection Defense

OpenAI has implemented safeguards to protect against malicious instructions hidden in web content that might try to manipulate ChatGPT's behavior. While no system is perfect, these defenses are continuously updated.

Agent Mode Restrictions

Agent Mode cannot:

  • Execute arbitrary code in your browser
  • Access your file system
  • Install extensions or modify browser settings
  • Access other applications on your computer

Sensitive Site Protection

Agent Mode pauses before taking actions on financial institutions and other sensitive sites, requiring your explicit approval to proceed.

Data Storage and Handling

What Data Does Atlas Collect?

Atlas collects different types of data depending on your settings:

Always Collected (Required for Functionality):

  • Your ChatGPT account information
  • Technical data (browser version, OS, performance metrics)
  • Error reports and crash data

Collected When You Use Features:

  • Chat conversations with ChatGPT (subject to your ChatGPT privacy settings)
  • URLs you visit (when ChatGPT visibility is enabled)
  • Browser memories (only if you enable this feature)

Optional (Opt-In Only):

  • Web browsing content for AI training (off by default)

Where Is Your Data Stored?

  • Local Storage: Browsing history, cookies, and cached data remain on your device
  • OpenAI Servers: Chat conversations and browser memories (if enabled) are stored in your ChatGPT account
  • Encryption: Data transmitted to OpenAI is encrypted in transit and at rest

How to Delete Your Data

  1. Clear Browsing History: Settings → Privacy → Clear Browsing Data
  2. Delete Browser Memories: Settings → Browser Memories → Archive or Clear All
  3. Remove ChatGPT Data: Through your ChatGPT account settings
  4. Complete Account Deletion: Contact OpenAI support to delete all associated data

Privacy Best Practices

Start Conservative

Begin with Browser Memories disabled and enable only when you understand the feature

Use Site Controls

Turn off visibility for sensitive sites like banking, healthcare, and personal email

Review Regularly

Periodically check your browser memories and archive ones you no longer need

Use Incognito

Browse in incognito mode when you want complete privacy from ChatGPT

Understand Training

Keep data training opt-out unless you specifically want to help improve the models

Monitor Agent Mode

Always watch what Agent Mode is doing, especially on logged-in sites

Transparency and User Rights

Your Rights

As an Atlas user, you have the right to:

  • Access: View all data associated with your account
  • Deletion: Request deletion of your data at any time
  • Portability: Export your browser memories and chat history
  • Opt-Out: Disable features or data collection you're uncomfortable with
  • Transparency: Understand how your data is being used

OpenAI's Commitments

OpenAI has publicly committed to:

  • Not using your browsing data for training by default
  • Respecting website owners' opt-outs (GPTBot)
  • Providing clear, accessible privacy controls
  • Continuous improvement of security and privacy features
  • Transparent communication about data practices

Important Privacy Considerations

While Atlas provides extensive privacy controls, users should be aware:

  • Browser memories are stored in the cloud and could theoretically be accessed by OpenAI employees with appropriate authorization
  • Like all cloud services, Atlas is subject to legal requirements and may need to comply with government requests
  • No system is perfectly secure; consider the sensitivity of data you share with any AI tool
  • Agent Mode, while protected, operates on websites as you and could theoretically be exploited through sophisticated prompt injection attacks

Continue Exploring

Getting Started Guide Back to Home