Anthropic turns its Chrome side panel into Claude Cowork

If you've ever used a browser sidebar AI, you know the frustration: you close the tab, the conversation disappears, and you have to re-explain the whole project when you switch back to your desktop app.

Anthropic just fixed that disconnect. The Claude in Chrome extension has been upgraded to a full Claude Cowork session, syncing browser tasks with Claude's desktop, web, and mobile apps. Your browser chats now live in your permanent account history instead of disappearing into the void.

The feature is available today for Max and Team subscribers, with a rollout to Pro users coming over the next few weeks. Enterprise plans have it turned off by default, though admins can enable it and restrict it to approved domains.

Browser automation meets cross-device workflow

Previously, Claude's Chrome side panel was an isolated sandbox. Now, it runs the exact same Cowork session as the rest of your Claude apps, meaning your custom skills and connectors work right inside the browser.

This is especially handy for bridging the gap between modern AI and legacy tools. Lots of internal company dashboards, old database front-ends, and vendor billing portals lack modern API integrations. Claude in Chrome bypasses that limitation by working directly through your existing browser logins, clicking links, typing, navigating across tabs, and filling forms on your behalf.

Anthropic offered a practical scenario: imagine building an expense report from multiple vendor portals. You can ask Claude in Chrome to open the tabs, pull dates and invoice totals, and format the spreadsheet. Later, you can jump over to the Claude desktop app to drop in local files from your computer or compare the numbers against last month's budget without losing any context.

That said, Claude in Chrome still doesn't support other Chromium-based browsers or mobile browsers, and you still need the dedicated desktop app if you want Claude interacting with local computer files.

Taming the prompt injection beast

Letting an autonomous AI click around the live web is inherently risky. Hidden text on malicious web pages, emails, or rogue documents can hijack an agent through prompt injection, tricking it into doing things you never intended.

To balance safety and usability, Anthropic introduced a smarter permission check. If you enable "automatically approve," Claude handles mundane multi-step tasks without nagging you at every click.

However, right before executing consequential actions, like submitting a form, sending an email, or downloading a file, an independent verification check compares the proposed action against your original prompt. If the action doesn't match what you actually asked for, it gets blocked. Highly sensitive actions, like making purchases or exposing personal data, still require manual human confirmation.

Why it matters

Browser sidebars have historically felt like temporary scratchpads. By turning its Chrome extension into a persistent Cowork session, Anthropic is making agentic workflows feel like a unified experience rather than a collection of fragmented tools.

For knowledge workers, this provides real, immediate utility. You don't have to wait for your IT department to build an API integration for every stubborn internal tool; Claude can simply navigate the web interface using your authenticated session. That dramatically lowers the barrier to automating repetitive administrative work.

We think Anthropic’s safety approach is particularly clever here. Requiring users to approve every single browser click causes alert fatigue, which ironically makes people approve bad actions blindly. Having a background monitor compare AI actions to original user intent gives you real autonomy with a guardrail. Still, Anthropic openly admits prompt injection cannot be 100% eliminated, so keeping Enterprise access off by default with strict domain whitelisting is a necessary, if cautious, compromise.