Until two weeks ago, letting Cowork use the web meant lending it your Chrome. On August 26 that changed: Claude now has a browser of its own, living in the Cowork side panel, with its own logins and none of your tabs. Since then Anthropic has also shipped Claude Fable 5.1 - the model Cowork runs at Medium effort - and a small Claude Code command that answers a question a lot of plugin collectors have never thought to ask.
Claude Now Has Its Own Browser in Cowork
Announced on August 26, Cowork in the Claude desktop app now has a browser built in. When a task needs a website, a browser opens in the side panel next to the task and Claude navigates pages, reads them, clicks, types, and fills forms while you watch. Links in the task transcript open in the same panel. There is no extension to install and no setup step: once it reaches you, giving Claude a task that involves a website opens the browser on its own.
The important detail is whose browser it is. It is Claude's, not yours. It is separate from the browser you use, and Claude does not see your tabs, bookmarks, or saved passwords. If you want Claude signed in to a site, you import cookies deliberately and site by site - from Chrome, Edge, or Firefox on macOS, and from Firefox on Windows and Linux. Banking, email, and single sign-on sites stay unchecked by default, and import is not available from Safari at all. You can also just sign in as you go, and Claude keeps those logins across Cowork sessions on that computer.
It is rolling out to Pro, Max, and Team plans in the Claude desktop app on macOS, Windows, and Linux, with Linux in beta. On Enterprise plans it is available now and admins manage it under Organization settings, then Cowork, then Built-in browser. One constraint worth knowing: the browser lives in the desktop app, so the app has to be open and online for Claude to use it - even when the Cowork session itself is running in the cloud. You can start a task on desktop and keep steering it from web or mobile, as long as the desktop app stays up.
Which Browser, and When
Cowork can now reach the web two ways, and Anthropic draws a clean line between them. The built-in browser is for handing a web task off while you keep working: gathering research for a report, or collecting this month's invoices from a vendor portal that has no connector. Claude in Chrome is for the page already in front of you, with the accounts you are already signed in to - updating your CRM, working through your inbox, editing the doc you have open.
If you already use Claude in Chrome, it keeps working and stays your default. If you do not have the extension, Claude uses the built-in browser once it reaches you. You can change it under Settings, then Cowork, then Preferred browser, and you can override it in a single task by naming the browser you want.
On safety, Anthropic is direct: the built-in browser carries the same prompt injection risk as any agent acting in a browser, where instructions hidden in a page try to redirect Claude. It runs the same safeguards as Claude in Chrome - permission before acting on a site the first time, high-risk sites blocked outright, and checks that compare each action against what you actually asked for. Those measures reduce the risk without eliminating it, and the guidance is to start on sites you trust. Anthropic also advises against using either browser to act on financial accounts, medical information, or other people's personal data.
Fable 5.1 Lands, and Cowork Runs It at Medium
On September 1, Anthropic released Claude Fable 5.1 and Claude Mythos 5.1. They are the same underlying model with different levels of safeguards: Fable 5.1 is generally available, while Mythos 5.1 is reachable only through trusted access programs for vetted cyberdefenders and life scientists.
For Cowork users the most concrete line in the announcement is a parenthetical: Fable 5.1 defaults to High effort in Claude Code, and to Medium in Claude Cowork and on Claude.ai. That is a deliberate difference, not an oversight - Anthropic's own charts show Fable 5.1 at Low or Medium effort matching or beating Fable 5 at considerably lower cost. Cowork is tuned for the shape of knowledge work rather than for long unattended coding runs.
The pricing change is the other headline. Input and output are unchanged at 10 dollars and 50 dollars per million tokens, but cache reads drop 75 percent to 25 cents per million. Anthropic estimates that works out to roughly 25 percent less than Fable 5 for typical workloads and up to about 45 percent less for context-heavy, tool-heavy agentic ones. Fable 5.1 is available to Pro, Max, Team, and Enterprise users, and to developers on the Claude Platform, AWS, Google Cloud, and Microsoft Azure.
Shipped alongside it: Enterprise Frontier Safeguards, an opt-in arrangement where enterprise data sits in cloud infrastructure the customer controls rather than Anthropic's, with any human review done by the customer by default - the privacy of a zero data retention agreement without giving up misuse detection. Anthropic says it was developed with more than 100 customers and will roll out in phases starting this fall, supported on Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Google's Agent Platform, and Microsoft Foundry.
A New Command Shows What Your Skills Cost
Claude Code 2.1.261, released September 4, added /skill-doctor. It reports which of the skills loaded into your session were never actually invoked, and what each one costs you in context - then tells you where to switch them off. In an interactive session the report opens in the plugin manager's Stats tab; run headless with the print flag and it comes back as text. It covers the skills in your session other than bundled and enterprise ones, and needs Claude Code 2.1.252 or later.
The reason it matters is the arithmetic underneath it. Every skill in the listing adds to your context on every single turn, whether or not Claude ever uses it, and the listing budget scales at one percent of the model's context window. Install enough plugins and you are paying that on every message of every session, for capability you may never invoke.
This is a Claude Code command, not a Cowork one. But the mechanism is identical wherever skills and plugins load, and Cowork users who have been collecting plugins since the enterprise expansions earlier this year are paying the same tax with no equivalent readout. Worth watching whether the report makes the trip across.
Why This Matters for Cowork Users
The web was the last real hole in Cowork's reach. Connectors covered the tools with APIs behind them; everything else - the vendor portal, the state licensing site, the internal tool nobody ever wrote an integration for - meant borrowing your own browser or doing it yourself. That hole is now filled, and filled in a way that does not require handing Claude your browsing session. Put it next to Drop #25's shared memory and the picture is a Cowork task that knows your context and can reach a page that has no API.
The honest counterweight is that reach and exposure grew together. A browser that can sign in to portals on your behalf is exactly what a prompt injection wants to get hold of. Anthropic's own framing is the right default, and it is unusually plain for a launch post: start on sites you trust, stay close to tasks with real consequences, and stop the task if something looks off.
Tip of the Week: Pick the Browser Before You Write the Prompt
Now that Cowork has two ways to reach the web, the choice between them is part of the task, not a setting you configure once and forget. The test is simple: does this job need your session, or just a session?
❌ Weak
"Go to the vendor portal and pull down last month's invoices."
Sent from your phone, with the desktop app closed, having never
decided which browser Claude should be using.✅ Strong
Settings, then Cowork, then Preferred browser. Choose the built-in
browser. Import the vendor portal login when it offers - leave
banking and email unchecked.
Then: "Open the vendor portal in your browser, sign in with the
login I imported, and download every invoice dated August."The rule of thumb: if the work is on a page already open in front of you, use Claude in Chrome. If you want to hand the web part off and keep working, use the built-in browser - and remember the desktop app has to stay open for it to run.
Worth Reading This Week
- → Claude gets its own browser in Cowork - Claude Blog - The August 26 announcement: the side-panel browser, why it is separate from yours, and how it differs from Claude in Chrome.
- → Use the built-in browser in Claude Cowork - Anthropic Support - The practical version: rollout by plan, importing logins site by site, safety, and the admin controls on Team and Enterprise.
- → Introducing Claude Fable 5.1 and Claude Mythos 5.1 - Anthropic - The September 1 release, including the cache-read price cut and the note that Cowork runs Fable 5.1 at Medium effort.
- → Enterprise Frontier Safeguards - Anthropic - Zero-data-retention privacy with misuse detection, using cloud storage the customer controls. Phased rollout starting this fall.
- → Extend Claude with skills - Claude Code Docs - Where the new skill-doctor command is documented, along with the context budget every loaded skill draws from.
- → Claude release notes - Anthropic Support - The running log of what shipped, worth checking before you assume a feature has reached your plan.
July gave Cowork more surfaces. August gave it memory that travels between them. This stretch gave it reach - a way to get to the parts of your work that never had an API. The interesting question from here is not whether Claude can get to the page, but whether you meant to send it there.
Until next time.
— The ClaudeCowork.com team