Claude in Chrome exits preview, acts on pages without asking each time
By Steven Van ·
General availability brings autonomous browser actions to all paid plans, gated by a safety classifier and stronger prompt-injection defenses.
Claude in Chrome has moved out of preview and is now generally available on every paid Claude plan. The extension lets Claude view the page you're on and take actions like reading and typing text, clicking links, navigating between pages, and filling out forms, using your existing logins, for tools that don't connect to Claude directly, such as internal dashboards, legacy systems, and vendor portals.
With general availability, Claude can now take actions autonomously in the browser instead of asking for approval before every one. A safety classifier checks each action against what you asked for before it runs, and blocks anything that doesn't match. You can still switch back to manual approval in settings.
Anthropic also says it has strengthened its defenses against prompt injection, where malicious instructions hidden in web pages, emails, or forms try to redirect the agent. In its latest evaluation, using attacks sourced by professional red-teamers, no attacks succeeded against Claude Sonnet 5 or Opus 5 when running with the probes and safety classifier, while Fable 5 had a 0.3% success rate.
- Install from the Chrome Web Store; it doesn't yet run on other Chromium browsers or on mobile.
- On Enterprise plans, admins can manage Claude in Chrome in Organization Settings and restrict it to approved domains.
- Working with files on your computer or other applications still requires the Claude desktop app.
