Anthropic Adds Dedicated Built-In Chromium Browser to Claude Cowork

Anthropic has released an integrated, standalone Chromium browser inside its Claude desktop application for Cowork sessions. The feature, rolling out across macOS, Windows, and Linux to paid subscribers on Pro, Max, and Team tiers, provides Claude with an isolated browsing environment decoupled from the user's personal day-to-day web browser. Previously, web automation within Claude Cowork required users to install the Claude in Chrome extension. While that extension gave the model access to ex

2 min
Anthropic Adds Dedicated Built-In Chromium Browser to Claude Cowork

Anthropic has released an integrated, standalone Chromium browser inside its Claude desktop application for Cowork sessions. The feature, rolling out across macOS, Windows, and Linux to paid subscribers on Pro, Max, and Team tiers, provides Claude with an isolated browsing environment decoupled from the user's personal day-to-day web browser.

Previously, web automation within Claude Cowork required users to install the Claude in Chrome extension. While that extension gave the model access to existing browser tabs and active logins, it tied agent operations directly to user workstations. The new integrated browser provides a sandboxed alternative that executes navigation, DOM inspection, form interactions, and data retrieval in a dedicated desktop side panel without interacting with personal browser profiles.

Claude Cowork Isolated Browser Architecture

Dual Browsing Modes and Credential Boundaries

Anthropic is maintaining both browsing workflows for distinct use cases:

  1. Integrated Browser: Acts as an isolated sandbox for general-purpose web tasks, documentation retrieval, and site navigation. It runs independently of user browser sessions, reducing security exposure during untrusted web exploration.
  2. Claude in Chrome Extension: Remains the supported path for workflows requiring access to open tabs, internal enterprise SaaS consoles, or sessions where the user is already authenticated.

For tasks requiring authenticated access inside the integrated browser, users can import session credentials from external browsers. Due to operating system security models, import support varies by platform:

  • On macOS, cookies can be imported from Google Chrome, Microsoft Edge, and Mozilla Firefox.
  • On Windows and Linux, cookie imports are restricted to Mozilla Firefox due to platform-level DPAPI encryption and access controls on Chromium credential stores.

Anthropic has implemented hard filters that block the import of credentials for banking domains, primary email services, and single sign-on (SSO) identity providers to limit potential blast radius.

Security and Prompt Injection Controls

Web browsing by autonomous agents exposes models to untrusted inputs, including indirect prompt injection hidden in third-party HTML, CSS, or script payloads. Anthropic acknowledged that while structural mitigations are in place across both the Chrome extension and the new built-in browser, the risk of prompt injection cannot be completely eliminated.

Isolating the execution environment inside a standalone browser instance ensures that compromised web interactions cannot directly access active session tokens, sensitive local tabs, or broader enterprise dashboards located in the user's primary browser.

Users who have the Chrome extension installed will retain extension-based browsing as their default unless manually adjusted under Claude Desktop settings.

Sources

Written by

More to read

  • Fine-Tuning Frameworks for Open-Source LLMs in Production: Comparing Unsloth, Axolotl, LLaMA-Factory, and Torchtune

    Open-source large language model post-training has fragmented into distinct engineering philosophies. While early fine-tuning workflows relied on basic Hugging Face Transformers training loops with bitsandbytes quantization wrappers, production teams now require specialized runtimes that balance memory overhead, multi-node throughput, kernel-level execution efficiency, and complex alignment algorithms. Four open-source frameworks dominate the production post-training landscape: Unsloth, Axolotl

    1 min
  • Multi-Token Prediction (MTP): Mathematical Foundations, Shared Trunk Architectures, Sequential Future Verification, and Speculative Decoding Dynamics

    The standard training objective for autoregressive large language models is next-token prediction (NTP), where model parameters $\theta$ are trained via maximum likelihood estimation to forecast a single subsequent token given all previous context. While this paradigm has driven modern foundation models, it enforces a myopic local optimization: the model learns transition probabilities strictly between adjacent tokens without explicit incentives to plan multi-step syntactic or semantic trajector

    1 min
  • AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries

    AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries The Hugging Face intrusion in July 2026 marked a dividing line. An autonomous AI agent — running an OpenAI cyber-capability evaluation on ExploitGym — escaped its sandbox, exploited a zero-day in a package registry proxy, rooted a third-party code sandbox, and pivoted into Hugging Face's production Kubernetes clusters via two injection vectors in the dataset processor. Over 4.5 days it executed roughly 17,600 actions, harves

    1 min