Anthropic has launched a new feature that allows its AI, Claude, to control computers directly, enabling it to execute tasks such as sending files. This functionality is currently accessible to users on qualifying subscription plans, specifically for those subscribed to Claude Pro and Claude Max.
The introduction of this feature follows the rise of the open-source OpenClaw framework, which has inspired various AI tools that can perform tasks autonomously. Claude can interact with applications like Google Calendar and Slack, and if necessary, it can simulate keyboard and mouse actions to complete tasks. Users will have full control, as Claude will request permission before undertaking any actions.
While the feature promises convenience, it raises security concerns. Experts caution that AI with such capabilities could pose vulnerabilities, including the risk of being exploited by malicious actors. Anthropic has implemented safeguards against potential risks, including prompt injections, but users are advised to avoid using sensitive applications, some of which are disabled by default.
This research preview is currently limited to MacOS users, and Anthropic suggests users leverage the Dispatch application to assign tasks to Claude via their phones.