Most browser AI tools feel like a separate brain bolted onto your actual workflow. You get help in the sidebar, then manually carry that context somewhere else. Anthropic is trying to fix that. The Claude in Chrome side panel has been renamed and rebuilt as Claude Cowork, and the core change is that your browser session is now the same session you use everywhere else. Anthropic announced the update on August 12, 2026, with availability on Max and Team plans today, and a rollout to Pro users coming in the next few weeks.
What actually changed
Before this update, conversations in the Chrome side panel were siloed. Nothing carried over to the desktop or mobile apps. Now, Claude Cowork runs as a single persistent session across surfaces. Start a task in the browser, pick it up later on desktop. That’s the pitch, and it’s a straightforward one.
The browser extension itself lets Claude interact directly with web pages, clicking links, filling forms, reading content, and navigating tabs using your existing logins. That makes it useful for tools that don’t have native Claude integrations, things like internal dashboards, vendor portals, or legacy systems. Anthropic gives a practical example: pulling invoice data from several vendor portals, having Claude open each tab, read the amounts, and populate a spreadsheet, then switching to the desktop app to merge it with last month’s budget file.
It’s a credible workflow. But it also illustrates who this is actually built for. This is not a casual use case.
The agent risk problem isn’t solved, just managed
Anthropic is straightforward about the risks here, which is worth noting. Claude in Chrome is an AI agent acting in a browser, and prompt injection is a real threat. Malicious instructions hidden in web pages, emails, or documents can redirect Claude to take actions you never intended.
Since the pilot, Anthropic has added a secondary review layer that checks Claude’s actions against your original request before anything consequential happens, such as submitting a form, sending a message, or downloading a file. The system also still prompts you before irreversible actions like purchases or sharing personal data. These are useful guardrails. But Anthropic is honest that they reduce risk, not eliminate it.
- Automatic approval mode lets Claude work through multi-step tasks without constant check-ins
- A secondary check blocks actions that don’t match the original request
- Manual confirmation is still required for purchases, data sharing, and other high-stakes actions
- Enterprise admins can restrict Claude in Chrome to approved domains only
How it fits the broader market
Microsoft’s Copilot has browser integration through Edge, and Google’s Gemini is baked into Chrome itself. Both have home-field advantages Anthropic simply doesn’t have. So Anthropic is competing on session continuity and agent capability rather than distribution. That’s a reasonable bet if Claude Cowork can consistently handle multi-step browser tasks without going off-script. For developers and ops-heavy teams dealing with fragmented tooling, the cross-surface session model is genuinely useful. It’s also a clear signal that Anthropic sees agentic, multi-step work as the product direction, not just chat.
