Brave tests AI browsing mode with opt-in feature flag, isolated profile & other safeguards

Brave has launched its experimental AI browsing feature in the Brave Nightly development channel. This release targets early adopters and invites user feedback to shape future improvements. While Brave's long-term goal is to evolve the browser into an intelligent assistant that automates web tasks and enhances user productivity, the company stresses the need for caution in deploying agentic AI.

As part of this approach, Brave warns that giving an artificial intelligence agent control over browsing introduces risks like data exposure, unintended actions, and indirect prompt injection attacks. These challenges affect all AI-powered browsers, prompting Brave to emphasize careful development and security community involvement.

To mitigate risks, Brave restricts initial access to opt-in testers. The AI browsing runs in a separate isolated profile, only activates via manual user invocation, and includes various usage restrictions as well as reasoning-based security defenses. These protections aim to limit the potential impact of agentic actions while Brave continues gathering feedback and refining the system.