Claude Cowork gets its own browser, deliberately separate from yours

Anthropic put a browser inside Claude Cowork on August 26. When a task needs a website, one opens in the side panel of the desktop app, and Claude navigates it, reads pages, fills forms and finishes the job there. The interesting part is not that it browses, but whose browser it is.

A separate browser is a design decision, not a limitation

Anthropic states the boundary plainly: the built-in browser is separate from the user's own, and Claude never sees their tabs, bookmarks or passwords. That is the opposite of the approach taken by agents that work inside the browser a person already uses, where the value comes precisely from the accounts that person is already signed in to.

Both designs are coherent, and they buy different things. An agent in your own browser can book and buy because your sessions are already there. An agent in its own empty browser cannot, and in exchange nothing it encounters on a page can reach a session it was never given. The cost lands on the user, who has to hand over whatever access a task genuinely needs, deliberately, instead of having it all present by default.

Where it runs and who gets it

The browser is part of the Claude desktop app, on macOS, Windows and Linux, with Linux still in beta. It went to Pro, Max and Team plans over the week beginning 26 August, and arrives switched on. Enterprise administrators can manage it from the day of the announcement.

This is the desktop app specifically, not the web version and not the browser extension, so it stands beside rather than replaces the browser tooling Anthropic already ships for developers through the API.