Anthropic just gave Claude two ways to browse the web. Claude Cowork on desktop now comes with its own standalone browser, opening in the sidebar whenever a web task pops up. Claude can flip through pages, click, type, and fill out forms on its own — no Chrome extension required. Meanwhile, Claude in Chrome has left beta and is now available to all paid plans.
The biggest difference? Login state. Cowork uses Claude's own browser, walled off from your everyday browser, so it can't see your tabs, bookmarks, or passwords by default. You can import login details on a per-site basis. Claude in Chrome, on the other hand, operates directly in the Chrome you've already opened and signed into — making it better suited for your inbox, CRM, or whatever page you're actively editing.
The Chrome version also relaxes the manual approvals. Previously, Claude basically waited for you to approve every step. Now it first scans the page for malicious instructions, then checks whether the action it's about to take matches your original task. If the safety checks pass, it keeps going autonomously. In Anthropic's latest red-team testing, neither Sonnet 5 nor Opus 5 saw a successful prompt injection attack, and Fable 5 succeeded only 0.3% of the time. Still, the company acknowledges that this kind of risk can't be completely eliminated.