Brave Introduces Agentic AI Mode for Automated Browsing Tasks

Brave, the privacy-focused browser, has unveiled a new AI browsing feature that utilizes Leo, its privacy-respecting AI assistant, to carry out automated tasks on behalf of users.

The feature, currently in the testing phase and accessible through the Brave Nightly version, is designed to aid users with tasks like autonomous web research, product comparisons, promo-code discovery, and news summarization.

This new agentic AI browsing mode, which is initially disabled by default, marks the initial step towards deeper AI-user integration for the privacy-centric browser.

Addressing AI Risks

Brave emphasizes that agentic AI browsing carries inherent risks and should not be used for critical operations due to the potential for prompt injection attacks and misinterpretation of user intent.

To mitigate these risks, the new mode operates on a separate, isolated profile that does not have access to sensitive user data such as cookies and login information.

Restrictions are in place to prevent access to certain areas, including the browser’s settings page, non-HTTPS sites, and the Chrome Web Store for extension downloads.

All actions taken by the AI browsing mode will be visible in tabs, with warnings issued to users for any risky behavior, requiring explicit user approval.

Furthermore, an ‘alignment checker’ mechanism will monitor the mode to ensure that the AI’s actions align with user intent, similar to Google’s approach with Gemini on Chrome.

Brave has implemented specific policy-based rules and trained models to combat prompt injection attacks, offering effective protection for users.

When it comes to data privacy, Brave guarantees that user data will not be used for AI model training, maintaining its core values of ad/tracker blocking and a strict no-logs policy.

Testing the New Mode

Individuals interested in testing Brave’s agentic AI mode can do so through Brave Nightly by enabling the “Brave’s AI browsing” flag in ‘brave://flags.’

Once activated, a button within Leo’s chat box will initiate the new browsing mode for testing purposes.

Feedback from testers is encouraged to help address any issues, with Brave increasing its HackerOne bug bounty payments for in-scope submissions related to AI browsing.

Broken IAM isn’t just an IT problem – the impact ripples across your whole business.

This practical guide covers why traditional IAM practices fail to keep up with modern demands, examples of what “good” IAM looks like, and a simple checklist for building a scalable strategy.

Get the guide