Anthropic now gives Claude its own browser on desktop computers. As the company announced Wednesday, Claude's desktop application for Mac, Windows, and Linux includes access to a built-in Chromium-based browser in Cowork that enables the AI assistant to navigate the web directly within the workspace. The update brings Anthropic's desktop offering in line with OpenAI, which recently abandoned plans for Atlas, its standalone browser product, and instead integrated browsing into the ChatGPT desktop application.
The new capability is rolling out to subscribers on Pro, Max, and Team payment plans. Before this release, users needed to install a Chrome extension for Claude to browse the web—an approach that functioned adequately but felt makeshift. The original Claude Chrome extension launched exactly one year earlier, on August 26, 2025. Just two weeks before Wednesday's announcement, Anthropic had pushed a major update to that Chrome extension, essentially converting it into a Cowork session. Users can import login credentials from Chrome, Edge, and Firefox on macOS, and from Firefox on Windows and Linux. Banking logins, email provider credentials, and single sign-on sites are explicitly excluded unless users deliberately choose to grant Claude access.
According to Anthropic, the Chrome extension still serves a purpose when users want Claude in the sidebar to discuss a page already open in their own browser—such as updating a CRM, working through an inbox, or editing a document. "Claude in Chrome is for the page you already have open, with the accounts you're already signed in to," the company explains. If users have the extension installed, it remains the default option for Claude's web surfing unless they explicitly switch their preferred browser in Claude Desktop settings. The company stresses this is "Claude's browser, not yours"—a separate environment without the user's login information by default.
Anthropic warns that prompt injection risk exists whenever an AI agent operates on the web. The company built safeguards into both the Chrome extension and the built-in browser, but acknowledges that "the risk is not zero." Because the browser typically won't be logged into many sensitive sites, the potential damage from any security failure should remain limited. The distinction between "a browser" for general web tasks and "your browser" for work on pages already open reflects Anthropic's attempt to balance convenience with security—keeping AI assistance accessible while reducing exposure of personal accounts. For now, desktop users can choose between the integrated browser for standalone research and the extension for contextual assistance, depending on whether the task requires existing login sessions or can operate independently. The industry is converging on a model where AI assistants don't need separate applications to access the web—they bring the browser with them, embedded in the tools where users already spend their time.