Key Highlights
- Anthropic has introduced a capability allowing Claude to remotely operate Mac computers for task completion
- Mobile users can submit instructions that prompt Claude to launch applications, navigate websites, and populate spreadsheets
- This functionality integrates with Dispatch, enabling task delegation to Claude through smartphones
- The company cautions that this 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, sparking competitive responses across the technology sector
Anthroplic has rolled out an innovative capability that enables its Claude artificial intelligence to assume control of computers for task execution. Monday’s announcement revealed this functionality is currently accessible to Claude Pro and Claude Max members.
This capability operates exclusively on MacOS systems. When users transmit instructions via their smartphones, Claude executes the requested operations on their desktop machines — launching software, conducting browser navigation, completing spreadsheet entries, or reorganizing files.
A demonstration from Anthropic displayed a scenario where someone behind schedule for a conference call requested Claude to convert a presentation into PDF format and include it with a calendar appointment. Claude fulfilled the request autonomously.
The system interfaces with platforms such as Google Calendar and Slack using integrated connectors. When specific connectors aren’t present, Claude can manually simulate keyboard and mouse interactions to accomplish objectives.
Claude consistently requests authorization before engaging with new applications. Users maintain the ability to terminate any operation at their discretion.
Integration With Dispatch System
This computer control functionality operates in conjunction with Dispatch, a mechanism Anthropic introduced previously within Claude Cowork. Dispatch facilitates ongoing communication with Claude across multiple devices and enables assignment of persistent tasks.
When combined, these capabilities can manage activities such as delivering morning summaries, executing code evaluations, or processing email correspondence automatically.
Anthroplic acknowledges that sophisticated operations may require multiple attempts to complete successfully. The organization characterizes this as an experimental release designed to collect user input and determine priority areas for enhancement.
Privacy Concerns and Recommendations
Anthroplic has implemented protective measures to minimize exposure, including automated detection systems for prompt injection vulnerabilities. Nevertheless, the company advises users against deploying this feature with applications managing confidential information — certain sensitive applications are deactivated automatically.
Cybersecurity professionals have raised broader questions regarding agentic artificial intelligence. These systems can execute substantial operations rapidly with minimal oversight. Additional concerns exist regarding malicious actors potentially compromising agents to obtain unauthorized access to private documents or infrastructure.
Anthroplic confirmed ongoing refinement of protective protocols as emerging vulnerabilities surface.
This release positions Anthropic in more direct rivalry with OpenClaw, an open-source system that achieved widespread attention recently. OpenClaw permits users to transmit instructions through WhatsApp or Telegram for local device execution. Nvidia recently unveiled NemoClaw, a commercial adaptation of this technology.
OpenAI entered this domain last month by recruiting Peter Steinberger, OpenClaw’s developer, to contribute to personal AI assistant development.
Currently, Claude’s computer control remains in experimental preview status on MacOS exclusively for Pro and Max subscription tiers.
