Anthropic has added a built-in browser to Claude Cowork. The new feature lets the AI assistant access the web without relying on the Claude in Chrome extension.
Cowork can now detect when a prompt needs web information. It automatically opens Anthropic’s custom browser in a side panel.
Anthropic has enabled the feature by default. Enterprise customers can already use it. The company will roll it out to Pro, Max, and Team subscribers this week.
Administrators and individual users can switch back to the Chrome extension through the settings.
Anthropic explained the change simply: “A lot of web tasks don’t need your browser, just a browser, and now Claude has one.”
The Built-In Browser Has Limited Access
Anthropic has placed several restrictions on the new browser.
The browser cannot automatically access logged-in services. These include email, banking, and other personal accounts.
Users must manually enter their credentials to access such services. As a result, the browser mainly retrieves public information.
It does not automatically use existing sessions from a user’s regular browser.
Built-In Browsers Could Change User Choice
The change is significant because Chrome handles close to 70% of global web traffic, according to Statcounter.
European regulators have spent years trying to reduce the influence of browser defaults.
Article 6(3) of the Digital Markets Act requires designated gatekeepers to show users a choice screen during first use. The screen must display major browsers, search engines, and virtual assistants.
European rules have also pushed Google to create room for competing assistants on Android.
However, an in-app browser does not trigger these requirements.
Anthropic is not a designated gatekeeper. Regulators also do not classify Claude as a core platform service.
Users do not see a browser choice screen because they never directly open the browser.
Prompt Injection Remains a Security Risk
Anthropic acknowledges that the built-in browser does not remove all security concerns.
Prompt injection remains a major risk for AI agents. Attackers can hide malicious instructions inside webpages.
Those instructions may manipulate an AI agent into taking unintended actions.
Anthropic says the built-in browser uses the same security guardrails as the Claude in Chrome extension.
The risk is not only theoretical.
TNW previously reported that Claude Cowork could escape its local virtual machine. The report also claimed that the system accessed credentials stored on a Mac.
AI Companies Are Moving Toward Built-In Browsers
OpenAI has already taken a similar approach.
The company shut down its standalone Atlas browser this month. It then integrated the technology into a Work mode with its own built-in browser.
Anthropic has also made more features part of the default experience.
Earlier, the company combined Claude chat and Cowork memory by default. Users can still disable the feature afterward.
The broader shift could change how people use the web. More users may interact with browsers inside AI assistants instead of opening a traditional browser themselves.
