Anthropic is testing a Chrome extension to make Claude work directly in your browser

Anthropic has launched a Claude extension for Chrome, enabling trusted users to instruct Claude to take actions directly within the web browser. The feature is initially available to 1,000 Max plan users as part of a limited research preview. While this early access is restricted, users can join the Claude for Chrome waitlist for a chance to participate in subsequent phases.

As part of the pilot, Anthropic will gradually expand access while focusing on strengthening safety measures and building confidence in the product. Early versions of Claude for Chrome have assisted with managing calendars, scheduling meetings, drafting email responses, processing expense reports, and testing new website features. However, the company notes that some vulnerabilities remain, especially regarding prompt injection attacks. Such attacks involve hidden malicious instructions on websites or in documents that could trick an AI assistant into performing unintended actions.

To address these risks, Anthropic is experimenting with site-level permissions, action confirmations, and additional mitigation strategies to safeguard users. During the initial research preview, Anthropic advises using the extension only on trusted sites and avoiding those involving financial, legal, or medical information.