Anthropic's Claude Cowork Now Runs Full Agentic Sessions in Chrome

Anthropic just turned its Chrome extension from a research assistant into a full-fledged autonomous worker. The extension previously handled basic browser tasks—reading pages, clicking, typing, filling forms. Now a complete Claude Cowork session runs directly in the side panel, bringing skills, plugins, and connectors to the browser for the first time without any additional setup.

The practical difference is significant. Claude can now pull metrics from an analytics dashboard, organize files in Google Drive, or log sales calls in Salesforce—then turn that work into Excel spreadsheets, PowerPoint decks, or written reports. That’s a meaningful shift from “find information” to “do the job,” and it’s available now on all paid plans through the Chrome Web Store. Sessions are tied to your account rather than your device, so work follows you across machines.

Anthropic built in a guardrail for purchases: Claude asks for confirmation before spending money. But the company is also blunt about the risk that comes with giving an LLM control of your browser. Hidden instructions on websites remain a real prompt-injection vector. Anthropic explicitly advises against using the extension with banking or health data. That warning matters more now, not less, because Cowork has more capabilities—more surface area for something to go wrong.

This move puts Anthropic in direct competition with OpenAI’s Operator and Google’s Project Mariner on the browser-automation front, but the integration with Cowork’s existing skills ecosystem is the differentiator. The question is whether users will trust an autonomous agent inside a browser that they’re also using for sensitive tasks. The confirmation prompt on purchases is a start, but prompt-injection defenses will need to keep pace with the tool’s expanding reach.