Claude Cowork Gets Its Own Browser, Separate From Yours
Anthropic gave Claude Cowork its own browser this week — a surface fully separate from a user’s personal browser or the Claude in Chrome extension, built so Claude can navigate websites, read pages, fill forms, and run web research independently, without needing the user’s own browser open at all. It’s rolling out now to Pro, Max, and Team plans on macOS, Windows, and Linux, with Enterprise admins able to turn it on immediately through organization settings. Anthropic drew a hard privacy line around it: “the built-in browser is separate from your own. Claude never sees your tabs, bookmarks, or passwords.” Users can selectively import saved logins from Chrome, Edge, or Firefox, but banking, email, and SSO sites stay excluded by default unless a user opts them in — and the same prompt-injection safeguards used in Claude in Chrome carry over, checking intended actions against user instructions before Claude executes them.
The split is deliberate: Claude in Chrome handles tasks on pages a user already has open, while the new Cowork browser is for autonomous, session-independent work — the kind of standing research or data-collection task a person would otherwise have to babysit. It can also be driven remotely from web or mobile clients as long as the desktop app stays open and online, which stretches Cowork’s reach past the machine it’s installed on.
It lands two days after Anthropic shipped unified memory across Claude chat and Cowork — context built over months of conversation now carries automatically into Cowork tasks, updates in real time rather than only after a session ends, and stays visible and editable in a Topics list, with hard exclusions (Social Security numbers, government IDs, criminal history, immigration status) that hold regardless of a user’s settings. Read together, the two releases are the same bet from different angles: Cowork becoming a persistent, semi-autonomous work surface rather than a chat window bound to one session — the detail worth tracking before scoping how much autonomy an enterprise rollout actually grants it.