Claude gains "computer use": can now autonomously control PC and browser

Anthropic has launched a new capability for Claude that allows the model to take control of your computer. "Computer use" is now available as a research preview for Pro and Max subscribers, marking a turning point in how AI assistants function in practice.

Instead of responding with text, Claude can now click, scroll, and navigate browsers and applications on behalf of the user. If direct integrations are not available, it falls back to human-like computer control.

What does this mean in practice?

Imagine asking Claude to book a meeting room, compare prices across three supplier websites, or fill out a form in an external system. Claude now does it directly, without you needing to open a browser yourself.

For CIOs and enterprise teams, the implications are significant. Many enterprise applications lack modern APIs, and computer use is a way to connect AI assistants to these systems without integration work. It opens the door to automating tasks that previously required humans at the screen.

The technology is still in research preview, and Anthropic is clear that it is not ready for production use in sensitive environments. The risk of unintended actions in complex GUIs is real.

Competitors including OpenAI (GPT-5.4 Computer Use) and Google DeepMind have similar capabilities in development. The race for "agentic AI" that can actually operate in the real digital world is in full swing.

Availability: Research preview for Claude Pro and Max subscribers from March 23, 2026.