Lawrence Jengar
Mar 23, 2026 22:21
Claude AI can now control your Mac directly – opening apps, navigating browsers, and running dev tools. Available in research preview for Pro and Max subscribers.
Anthropic just gave Claude the keys to your computer. The AI company announced Monday that Claude can now directly control macOS machines – clicking, scrolling, and navigating applications to complete tasks while users step away.
The feature, launching in research preview for Claude Pro and Max subscribers, pairs with Dispatch, a mobile tool released last week that lets users assign tasks to Claude from their phones.
What Claude Can Actually Do
When Claude lacks a direct integration with a service, it falls back to controlling your screen like a human would. The AI opens files, operates browsers, and runs development tools automatically. Need a morning briefing compiled while commuting? Claude can pull emails and metrics using your actual applications.
For developers, the implications are more concrete. Claude can make changes in your IDE, run tests, and submit pull requests – all while you handle other work.
Anthropic built the system with a permission-first approach. Claude requests access before touching new applications, and users can halt operations at any point. The company also implemented automatic scanning to detect prompt injection attempts, a common attack vector for AI systems with computer access.
The Limitations Are Real
Anthropic isn’t overselling this. “Computer use is still early compared to Claude’s ability to code or interact with text,” the company acknowledged. Complex tasks sometimes require multiple attempts, and screen-based control runs slower than direct API integrations.
Some applications remain off-limits by default. Anthropic recommends avoiding sensitive data and sticking to trusted apps during the research preview. The company published safety guidelines but noted that “threats are constantly evolving.”
Broader Context
This release follows a busy period for Anthropic. The company shipped Claude Opus 4.6 on February 5 and Claude Sonnet 4.6 on February 17, both targeting complex agentic workflows – exactly the kind of tasks that benefit from computer control.
The timing also comes after Anthropic’s February clash with US federal agencies. The company refused to remove contractual prohibitions against using Claude for mass surveillance and autonomous weapons, leading to a phase-out of government usage.
Computer control represents the next frontier in AI assistants – moving from conversation partners to autonomous agents that operate in the real world. OpenAI and Google have signaled similar ambitions, but Anthropic shipped first on desktop.
For now, the feature requires macOS with the desktop app running. Mobile users can assign tasks through Dispatch, but the actual computer control happens on the Mac. Anthropic plans to learn from early usage before expanding platform support.
Image source: Shutterstock
