Key Points
- Anthropic introduced a capability allowing Claude to operate Mac computers for remote task completion
- People can delegate assignments from mobile devices while Claude handles applications, web navigation, and spreadsheet management
- This capability integrates with Dispatch, enabling task delegation to Claude through mobile interfaces
- Anthropic cautions that the technology remains experimental and recommends keeping it away from applications containing confidential information
- This development follows the viral success of OpenClaw, an alternative agentic AI system that sparked competitive momentum across technology companies
Anthropic revealed a capability that allows Claude AI to assume control of computers for task execution. The company made the announcement on Monday, with immediate availability for Claude Pro and Claude Max members.
The functionality currently supports MacOS exclusively. People transmit instructions from mobile devices, prompting Claude to execute desktop operations — launching applications, conducting browser sessions, populating spreadsheets, or organizing files.
Anthropic demonstrated the system through a scenario where someone behind schedule for a meeting requested Claude to convert a presentation into PDF format and add it to a calendar appointment. Claude fulfilled the request autonomously.
The system interfaces with applications like Google Calendar and Slack using integrated connectors. When connectors are unavailable, Claude manually operates keyboard and mouse functions to accomplish objectives.
Claude requires authorization before entering any new application. People maintain the ability to terminate operations at any moment.
Integration With Dispatch Functionality
The computer control feature operates in conjunction with Dispatch, a system Anthropic introduced recently within Claude Cowork. Dispatch enables ongoing communication with Claude across devices while assigning continuous responsibilities.
Combined, these capabilities manage activities such as delivering morning summaries, executing code validation, or monitoring email correspondence automatically.
Anthropic acknowledges that complicated assignments may require multiple attempts. The organization describes this as an experimental release designed to collect user feedback and pinpoint priority areas for enhancement.
Protection Measures and Cautions
Anthropic implemented protective measures to minimize vulnerabilities, including automated detection of prompt injection threats. However, the company continues to advise people against deploying the feature with applications managing confidential information — certain applications are deactivated by default.
Security professionals have raised wider concerns regarding agentic AI systems. These technologies can execute substantial actions rapidly with minimal advance notice. Additional risks include malicious actors commandeering the agent to obtain personal documents or system access.
Anthropic committed to continuous updates of protective measures as emerging threats develop.
Competitive Landscape Developments
This release positions Anthropic in closer competition with OpenClaw, an open-source system that gained widespread attention this year. OpenClaw permits people to transmit tasks through WhatsApp or Telegram for local device execution. Nvidia recently launched NemoClaw, an enterprise adaptation of the system.
OpenAI entered this domain last month by recruiting Peter Steinberger, OpenClaw’s creator, to develop personal AI assistant technology.
Currently, Claude’s computer control feature remains available as an experimental preview on MacOS for Pro and Max subscribers.
