Claude in Chrome Reaches General Availability
Claude CodeAnthropic moved Claude in Chrome out of beta and into general availability, giving all paid Claude Code plans (Pro, Max, Team, Enterprise) access to a browser extension that lets Claude see, click, type, and navigate the web directly. The extension now pairs with Claude Code's terminal workflow so developers can build in the CLI and verify the result live in the browser, including reading console errors, network requests, and DOM state. Anthropic also shipped safety hardening alongside GA, citing a drop in successful prompt-injection attacks from 23.6% to 11.2% after adding new defenses.
Featured Video
A video we selected to help illustrate this changelog
Key Takeaways
- General availability means all paid Claude Code plans (Pro, Max, Team, Enterprise) can now use Claude in Chrome without a beta waitlist.
- Prompt injection defenses cut the success rate of malicious in-page attacks from 23.6% down to 11.2%, though risk isn't eliminated.
- Build-test-verify workflow links Claude Code in the terminal with Claude in Chrome for live browser verification, including console and network inspection.
- Workflow recording lets users teach Claude a repeatable browser task once, rather than re-prompting it every time.
- Tool loading is now batched, replacing the prior one-request-per-tool pattern for faster session startup.
- Pro plan users are capped at the Haiku 4.5 model for browser automation according to early reviewers, meaning capability varies notably by subscription tier.
Sources & Mentions
4 external resources covering this update
General Availability
Claude in Chrome has exited its multi-month beta and is now generally available to all paid Claude Code plans. The extension turns the Chrome browser into an interface Claude can operate directly: it can see what's on screen, click buttons, fill in forms, and navigate between pages on the user's behalf.
Build-Test-Verify With Claude Code
The headline workflow pairing GA brings is a tighter loop between Claude Code's terminal agent and the browser. Developers can build a feature with Claude Code, then hand off to Claude in Chrome to test and verify the change live in the browser. Claude can read console output, inspect network requests, and check DOM state directly, closing the loop between "the code compiles" and "the feature actually works" without the developer manually re-verifying every change.
Workflow Recording
Claude in Chrome also supports teaching Claude a workflow by recording the steps once; Claude then learns to repeat that same sequence of actions for future, similar tasks. This is aimed at repetitive browser tasks that follow the same pattern each time, such as filling out the same form across many records.
Safety Hardening
Browser-operating AI agents are a known target for prompt injection, where malicious instructions are hidden in a webpage to trick the agent into unintended actions such as deleting files or making unauthorized purchases. Anthropic reported that its added defenses cut the success rate of such attacks from 23.6% to 11.2% in testing, roughly a 1-in-9 residual risk rather than 1-in-4. This safety work appears to be a major factor behind the extension staying in beta as long as it did.
Performance
Alongside the GA launch, Claude in Chrome's tool loading was also optimized: browser tools now load in a single batched request instead of one call per tool, which should make session startup noticeably snappier.