Anthropic adds built-in browser to Claude Cowork, ending extension dependency

Claude's desktop app now ships with its own Chromium-based browser inside Cowork, letting the AI assistant navigate the web independently — a move that puts Anthropic on par with OpenAI's browser-equipped ChatGPT desktop app.

Anthropic has rolled out a built-in browser for Claude Cowork on its desktop app, giving the AI assistant the ability to open webpages, read content, click buttons and fill out forms without relying on a separate browser extension. The feature is available on macOS, Windows and Linux, and is currently rolling out to Pro, Max and Team plan subscribers, with Enterprise administrators able to switch it on for their organisations.

From extension to built-in tool

Previously, Claude's only way to interact with the web inside Cowork was through the Claude in Chrome extension, which required hooking into a user's own browser session. According to Anthropic, that route remains useful for tasks tied to a page a person already has open — but many web-based chores don't need the user's browser at all, just a browser, which is what prompted the new built-in option.

The browser opens in a side panel within the desktop app whenever a task calls for one, allowing Claude to independently pull data from a dashboard, submit a form, or gather information from a site that lacks a dedicated connector — all while the user works on something else.

Kept separate from personal browsing

Anthropic has emphasized that this is "Claude's browser, not yours." The built-in browser runs isolated from a person's everyday browsing environment, meaning Claude cannot see existing tabs, bookmarks or saved passwords by default. Users who want Claude to access specific logins can transfer them selectively from Chrome, Edge or Firefox on macOS, or from Firefox on Windows and Linux — though banking, email and other sensitive single sign-on accounts are excluded from this option.

Anthropic also acknowledged that letting an AI agent act autonomously on the web introduces prompt-injection risks, where hidden instructions embedded in a webpage attempt to manipulate the assistant into taking unintended actions. The company said it has built in safeguards similar to those used in the Chrome extension, while cautioning that the risk "is not zero" and advising users to stick to trusted sites, particularly since the built-in browser typically won't be logged into sensitive accounts.

If a person already uses the Claude in Chrome extension, it will remain their default browsing option unless they change the preference in Cowork settings.