AI browsing in Brave now available for early testing (brave.com)

Brave has launched an early testing phase for its new AI browsing feature in the Brave Nightly build, designed to create a more intelligent browsing experience. This agentic browsing capability aims to automate tasks like researching, comparing products, and summarizing news. However, the feature also raises significant security concerns, as giving AI control over browsing can expose users to personal data risks and unintended actions. Brave is implementing a cautious rollout by keeping AI browsing behind an opt-in feature flag and soliciting feedback from security researchers to enhance safety. To address potential vulnerabilities, Brave has introduced several safeguards: AI browsing operates in an isolated profile to protect user data, incorporates reasoning-based defenses against malicious websites, and requires manual invocation by users. These precautions help mitigate the risks associated with indirect prompt injections and misinterpretation of user commands. Despite the challenges, Brave remains committed to refining the feature through user feedback, believing that a secure, intelligent browser can revolutionize the online experience while maintaining its core promise of privacy and security.