Anthropic Grants Claude Direct Computer Control in New Preview
Anthropic launches a research preview that allows Claude to navigate computers like a human, clicking and scrolling to complete tasks autonomously. The feature currently supports only Mac computers and requires explicit user permission for safety.
Anthropic introduces a new "computer use" feature for Claude that allows the AI assistant to control a user's computer directly. Available as a research preview for Claude Pro and Max subscribers, this functionality enables the AI to click, scroll, and navigate through web pages and applications. It pairs seamlessly with Dispatch, a recently launched mobile tool that lets users assign tasks to Claude from their smartphones for the AI to execute on their computers.
When Claude receives a task, it first checks for direct integrations like Google Calendar or Slack, but it falls back to visually navigating the screen like a human if those connectors are unavailable. This capability proves especially useful for developers, as Claude autonomously makes changes within integrated development environments, submits pull requests, and runs tests. Anthropic emphasizes a permission-first safety approach, meaning Claude requests access before interacting with any new application and users retain the ability to halt the AI at any time.
Anthropic openly admits the feature remains a work in progress and is not yet ready for widespread deployment. A significant limitation is that the tool currently only supports Mac computers, leaving Windows and Linux users waiting for future updates. Furthermore, screen-based operations run much slower than direct API integrations, and complex tasks often require multiple attempts to complete successfully, highlighting the ongoing challenges of granting an AI direct access to a user's system.