Anthropic Integrates Native Web Browser Into Claude Cowork

Artificial intelligence startup Anthropic has integrated a dedicated web browser directly inside its Claude Cowork platform, allowing the AI assistant to browse the internet within an embedded side panel rather than relying on an external extension for Google Chrome.

The native browser feature is activated by default for Enterprise tier users and is scheduled to roll out to Pro, Max, and Team subscribers throughout the week. Account administrators and individual users retain the option to manually revert back to using the Chrome browser extension inside their account settings. In an official statement regarding the release, Anthropic stated, "A lot of web tasks don't need your browser, just a browser, and now Claude has one."

To limit security exposure, the embedded browser operates with deliberate structural restrictions. The system does not automatically access stored user credentials for personal services such as email accounts or financial institutions unless credentials are manually keyed in by the user. Consequently, the utility functions primarily as a tool for retrieving publicly accessible web information rather than executing actions as an authenticated user.

The strategic shift bypasses dominant web access points at a time when Google Chrome commands nearly 70% of global web traffic, according to data from analytics firm Statcounter. By containing web browsing within the application interface itself, software providers can direct web tasks through internal environments rather than routing traffic through third-party desktop browsers.

This structural change also avoids regulatory mechanisms established under the European Union's Digital Markets Act. Under Article 6(3) of the DMA, designated gatekeeper companies are legally required to display choice screens presenting users with alternative search engines, web browsers, and virtual assistants. Because Anthropic has not been classified as a gatekeeper firm and Claude does not hold a core platform service designation, in-app web tools operate outside the jurisdiction of these mandatory choice prompts.

Security challenges around autonomous AI web access remain unresolved despite the internal integration. Anthropic acknowledged that prompt injection attacks—wherein malicious instructions hidden on external web pages compromise the AI agent's instructions—continue to pose a threat to the embedded browser, which utilizes the same protective safeguards as the previous Chrome extension. As first reported by The Next Web, previous security evaluations of Cowork identified vulnerabilities where the tool could breach its local virtual machine container to read user credentials stored on Mac systems.

Anthropic's move reflects a broader pattern across the generative AI sector as developers prioritize integrated default experiences. Earlier this month, competitor OpenAI decommissioned its standalone Atlas browser project in favor of incorporating a native browser into a unified Work mode. The deployment also mirrors Anthropic's existing strategy of enabling features by default, similar to its previous integration of Claude chat and Cowork shared memory systems.