Is the Cowork built-in browser meant to replace Claude in Chrome?
No — it's filling in an option that didn't exist before, not replacing an existing one. Before this update, if you wanted Claude to handle web tasks inside Cowork, the only path was through the Claude in Chrome extension, handing your own browser over to Claude. But a lot of web tasks don't actually need "your" browser — they just need "a" browser: looking something up on a public site, filling out a form that doesn't require login. That's exactly the gap the built-in browser fills: no extension to install, no need to use the browser you're currently working in — Cowork now has its own independent browser to use instead.
If you're already using Claude in Chrome for work on "the page you currently have open," official documentation notes that Claude in Chrome remains the right choice for that scenario — it isn't being displaced just because the built-in browser launched.
Why did Anthropic deliberately design the Cowork browser to be fully isolated from your own browser?
The most direct reason is a privacy boundary. If the Cowork built-in browser directly reused the tabs, bookmarks, and login state from your regular browser, it would mean Claude could, in theory, encounter private information unrelated to whatever task it's running — an open personal inbox, a social account you haven't logged out of. Fully separating the two draws a clear line: what Claude can access is only the login information you deliberately bring over, one site at a time — not everything sitting in your browser.
This design choice also explains why banking, email, and single sign-on sites are excluded from what can be imported — these are the categories where account misuse carries the most severe consequences, so the system doesn't offer that option even if a user actively wants to bring them over.
If I have a web task to hand off to Claude today, how do I actually decide which browser to use?
The most practical criterion is: where does the login state for this page currently live? If you already have the page open, already signed in, and the task is specifically about "this page" — editing an order you're viewing, replying to an email you're reading — Claude in Chrome is the obvious choice, no real decision needed. You don't need to bring login information over to a different browser; reusing the current state is simply the path of least effort.
If the work spans several different sites, you haven't logged into them yet, or it's something you'd rather hand off entirely and go do something else — the built-in browser is the better fit. You just import login information for the sites you'll need once, and Claude can then complete the work independently in its own browser without occupying the tab you're currently working in.
If a task involves both "the page you're currently looking at" and "several other external sites," the two can work together: let Claude in Chrome handle the update on the page in front of you, while the built-in browser goes and gathers data elsewhere — the two divide the work without conflict.
Is this built-in browser available to everyone right now? What limitations should I know about first?
It's currently in a staged rollout: the desktop version started rolling out to Pro, Max, and Team plan users this week; on Enterprise, availability depends on whether your organization's admin has turned it on — officially, admins can enable it starting today. The built-in browser currently only exists inside the Cowork desktop app; if you use Claude through the web or mobile apps, this feature isn't available to you yet.
Before actually using it, one thing worth confirming: if a task involves sensitive content you don't want Claude to accidentally encounter, the risk of prompt injection still exists even though the two browsers are isolated from each other — this isn't a risk unique to the built-in browser, it's a shared challenge for any AI agent that can act autonomously on the web. Starting with sites you trust and are familiar with, and watching how Claude actually operates, is a more solid approach than handing it an unfamiliar, complex task right out of the gate.
On August 26, 2026, Claude Cowork's desktop app added a built-in browser — meaning there are now two ways to let Claude work with web pages for you: the existing Claude in Chrome (a browser extension), and this newly launched browser built directly into the Cowork desktop app. The names sound similar, but what each is actually for, and which one you should pick, is what this article sorts out.
Claude in Chrome uses your browser — it's an extension installed on your Chrome, able to see the tabs you currently have open and directly reuse your existing logged-in sessions to take action. The Cowork built-in browser is the opposite: it's Claude's own browser, fully separated from the one you normally use — Claude can't see your tabs, bookmarks, or passwords. If a task needs a login for a particular site, you bring that login over one site at a time (currently supporting import from Chrome, Edge, or Firefox on macOS, and Firefox on Windows and Linux), while sensitive sites like banking, email, and single sign-on (SSO) are excluded by default.
Official documentation notes that when Cowork runs a task, it picks the most precise tool available, in order: first, connectors (Gmail, Google Drive, Slack) — if a ready-made connector exists, it's the fastest and most reliable path. Second, the browser — either the new built-in one, or your own Chrome via Claude in Chrome. If neither of those is available, Claude falls back to computer use, interacting directly with your screen by clicking, typing, and navigating apps. The logic behind this order is straightforward: pulling messages through a Slack connector takes seconds, but having Claude navigate Slack through your screen instead is much slower and more error-prone.
If the web task involves a page you already have open and are already signed into — updating a CRM record you're currently viewing, handling the email in front of you, editing a document you're already looking at — Claude in Chrome makes more sense, since it directly reuses your current login state without needing an extra import step.
If it's work Claude can complete independently, without you needing to stay and watch — pulling invoice amounts and dates from several vendor portals to build a budget spreadsheet, or filling out a form on an old system with no API — the built-in browser is the better fit, since you can hand the task off and keep using your own browser for something else at the same time, with the two staying out of each other's way.
Both browsers face the same core risk: prompt injection, where a malicious web page hides instructions attempting to trick Claude into taking an action you never asked for. Anthropic's documentation notes directly that these safeguards meaningfully reduce the risk but can't eliminate it, and recommends starting on sites you trust. The built-in browser is currently in a rolling rollout: it's available on desktop for Pro, Max, and Team plans, and Enterprise admins can turn it on for their organization starting today.