Anthropic's Claude AI can now remotely control computers through mouse clicks and keyboard inputs, letting users assign tasks from their phones and return to completed work on their desktops. The new capability in Claude Code and Claude Cowork tools allows the AI to point, click, and handle screens autonomously when direct app integrations aren't available.
Claude Pro and Max subscribers can access the feature as a research preview limited to macOS devices for now. The system requires users to run the Claude desktop app on a supported Mac, with pricing set at $20 per month for Pro and $100 monthly for Max plans.
When assigned a task, Claude first attempts to use existing connectors with services like Slack or Google Calendar. If those direct integrations aren't available, the AI falls back to controlling the computer like a human would, scrolling through interfaces, clicking buttons, and typing text using simulated mouse and keyboard inputs.
Anthropic says Claude will "always ask for your explicit permission" before exploring screens or clicking elements needed to complete tasks. The system uses a permission-first approach that requests user access before touching new applications.
The computer control capability pairs with Dispatch, a cross-device feature released last week that lets users assign Claude tasks from iPhones. This combination enables scenarios where someone could ask Claude via phone to export a PDF from their desktop computer and attach it to a calendar invitation while they're away from their machine.
Claude's version of computer control represents Anthropic's response to OpenClaw, an open-source AI agent framework that went viral earlier this year. Unlike OpenClaw's more open approach, Anthropic's implementation includes additional safeguards designed to minimize risks like prompt injections.
The company warns that "computer use is still early compared to Claude's ability to code or interact with text," noting that complex tasks might require second attempts. Anthropic also cautions that while training prevents "risky operations" such as moving money or modifying sensitive files, these safeguards "aren't perfect" and "aren't absolute."
Research preview availability began March 24 for qualifying subscribers who want early access to what Anthropic describes as its most autonomous AI capabilities yet.
