Anthropic has given Claude a genuinely independent way to browse the web, building a Chromium-based browser directly into Claude Cowork’s desktop app rather than continuing to route web tasks through the Claude in Chrome extension. Announced on August 26, the update means that when a task in Cowork requires visiting a website, Claude can now open its own browser in a side panel, navigate pages, click buttons, fill out forms and read content, all without touching the tabs, bookmarks or saved passwords in whatever browser you happen to use day to day.
Until this update, giving Claude web access inside Cowork meant granting it access to your actual browser through the Claude in Chrome extension, an approach that worked reasonably well but effectively required handing over your personal browsing session, tabs and all, whenever you wanted Claude to complete a task involving a website. Anthropic’s own explanation for the change is fairly direct, acknowledging that when the work involves a page you already have open, using your existing browser still makes sense, but that a substantial share of web tasks don’t actually need your personal browser at all, they just need a browser, and now Claude has one built in.
The built-in browser operates as a fully separate environment from your everyday browsing. Claude cannot see your open tabs, saved bookmarks or stored passwords in your personal browser, and nothing from your regular browsing session gets shared unless you deliberately choose to bring it over. Users who want Claude signed into specific websites can transfer logins one site at a time from Chrome, Edge or Firefox on macOS, or from Firefox on Windows and Linux, giving people granular control over exactly which accounts Claude can access rather than an all-or-nothing sharing arrangement. Sensitive categories like banking, email and single sign-on services are excluded from this transfer process by default, requiring users to make a deliberate choice if they want Claude to have access to those particular accounts, a design decision that reflects the genuinely elevated risk profile those account types carry compared to something like a vendor invoice portal.
This distinction between the two browsing modes matters practically for how people will actually use the feature going forward. The built-in browser is designed for handing off web-based chores entirely, tasks like researching material for a report, pulling this month’s invoices from a vendor portal, or working through a business system that lacks both an API and a dedicated Claude connector, letting users keep working on something else entirely while Claude handles the web portion independently. Claude in Chrome, by contrast, remains the better tool for situations where you’re already on a specific page, signed into a specific account, and want Claude to jump in and assist directly with what’s already open in front of you, such as updating a CRM record, working through an email inbox, or editing a document currently on screen.
The rollout schedule varies by plan tier. Enterprise customers already have access, with organization administrators able to manage the feature through Organization settings under Cowork and Built-in browser. The capability is rolling out this week to Pro, Max and Team plan subscribers on the desktop app, covering macOS, Windows and a Linux beta, giving the vast majority of Claude’s paying user base access within a relatively short window of the initial announcement. Notably, the browser arrives switched on by default for users on these plans, though anyone who prefers the previous Chrome extension approach can revert that preference in Settings under Cowork at any point.
Anthropic has been transparent about the security considerations that come with any AI system capable of independently browsing and interacting with websites. The company has explicitly flagged prompt injection risk as a genuine concern, the possibility that malicious instructions hidden within a webpage’s content could attempt to manipulate Claude’s behavior while it’s navigating that site, and has advised users to stick to trusted websites when relying on the built-in browser for autonomous tasks. That warning reflects a broader challenge facing the entire category of browser-using AI agents currently emerging across the industry, where the same capability that makes these tools useful, autonomously reading and acting on arbitrary web content, also creates a genuinely new attack surface that developers are still working out how to fully secure.
The timing of this launch places Anthropic in interesting company within the broader AI industry. OpenAI had previously pursued its own standalone browser project called Atlas before ultimately abandoning that approach and instead integrating browsing capability directly into the ChatGPT desktop app, a trajectory that closely mirrors what Anthropic has now done with Claude and Cowork. Rather than building and maintaining an entirely separate consumer browser product, both companies appear to have converged on the same conclusion, that embedding a purpose-built browsing capability directly into an existing AI assistant interface makes more practical sense than competing head-on in the standalone browser market against established players like Chrome, Safari and Edge.
There’s also a notable regulatory dimension to how this rollout has unfolded, particularly across European markets. Because the built-in browser ships switched on by default rather than presenting users with an explicit choice, some observers have pointed out that this default-on approach sidesteps the kind of browser choice screens mandated under the European Union’s Digital Markets Act, a regulation that specifically applies to companies designated as gatekeepers under that law. Anthropic does not currently carry that gatekeeper designation, meaning the company faces no legal obligation to present users with an explicit choice screen before enabling the feature, a distinction that has drawn some commentary about how quickly default-on AI features are becoming standard practice across the industry even as regulators focus their most stringent requirements on a narrower set of large platform companies.
For Claude users managing recurring web-based workflows, this update effectively removes what had been a somewhat clunky requirement, installing and configuring a browser extension just to give an AI assistant the ability to interact with websites on your behalf. With the built-in browser now handling that functionality natively inside the desktop app, Cowork moves closer to functioning as a genuine autonomous web-operating agent rather than a purely conversational assistant confined to text-based tasks, a shift that fits squarely within the broader industry trend toward AI systems capable of independently completing multi-step tasks across the open web rather than simply answering questions about it.
Full details on the built-in browser and how to configure it are available through Anthropic’s official announcement. For more coverage of AI agents and enterprise automation tools, visit Business Tech.
