Claude Cowork can now browse the web and do tasks for you
Claude just learned to use the web. Anthropic's new built-in browser lets AI read pages, click, type and complete tasks inside the desktop app all by itself!
Claude just got a little more hands-on. Anthropic has introduced a built-in browser for Claude Cowork, giving the AI assistant a dedicated way to work with websites from inside the Claude desktop app.
Claude AI can now open webpages, read information, click buttons, type into fields and fill out forms without using a person’s everyday browser. Here's all that you need to know!
Claude can browse on its own
The new browser appears in a side panel inside the desktop app. When a task needs a website, Claude can open the page and interact with it directly. That could mean collecting figures from a dashboard, downloading invoices from a vendor portal, filling out a form or gathering information for a report.
There is an important distinction, though: this is Claude’s browser, not the user’s.
Anthropic says Claude cannot see a user’s existing tabs, bookmarks or passwords. Users can choose to transfer logins from Chrome, Edge or Firefox on macOS, and Firefox on Windows and Linux, on a site-by-site basis.
Sensitive categories such as banking, email and single sign-on are excluded unless the user chooses to include them.
So what happens to Claude in Chrome?
Claude already has a browser extension called Claude in Chrome, and it is not going away. The extension is useful when users want Claude to work directly on a webpage already open in their browser. That could include editing a document, managing an inbox or updating a CRM.
The built-in browser is aimed at a slightly different job. It allows Claude to handle web tasks independently while the user gets on with something else. Users who already have Claude in Chrome will continue to use it by default, but they can change their preferred browser in Cowork settings.
There are still some safety questions
The browser is rolling out this week to Pro, Max and Team users through the Claude desktop app on macOS, Windows and Linux. Linux support is currently in beta. Enterprise administrators can also enable the feature for their organisations through Cowork settings.
But giving an AI the ability to click and type on websites comes with a new risk: prompt injection. This is when hidden or misleading instructions on a webpage try to manipulate an AI into taking an action the user did not request.
Anthropic says the built-in browser uses safeguards from Claude in Chrome, including checks that compare Claude’s actions with the original request. However, the company says these protections reduce the risk rather than eliminate it.
So starting with trusted websites is still a good idea.
Claude is moving from answering to doing
The bigger change here is not really the browser itself. It is what the browser allows Claude to do. Instead of telling users how to complete a task online, Claude can increasingly take care of the task itself.
That makes Cowork feel less like a chatbot and more like a digital assistant that can work across the web. And that is probably where AI assistants are heading next: less “here’s how”, more “I’ve done it”.