Claude Can Now Control Your Computer
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic launched a 'Computer Use' research preview for Claude, letting the AI browse the web, open files, and type or move the cursor as a human user would.
- The feature is available to Claude Pro and Claude Max subscribers on MacOS and Windows, with Claude requesting permission before each task and letting users stop it mid-action.
- Claude connects to apps like Google Calendar and Slack via integrations, and falls back to manually controlling the cursor and keyboard when no connector exists.
- Anthropic paired the feature with Dispatch, enabling users to assign tasks from their phone — such as creating a morning briefing or running tests while away.
- Anthropic disabled apps handling sensitive data by default and warns the feature is new and may contain errors, despite built-in safeguards against prompt injections and automatic vulnerability scanning.
- The launch follows the viral OpenClaw agentic AI tool and Nvidia's NemoClaw reference stack, with experts warning agentic AI can execute dramatic actions with little warning and be hijacked by malicious actors.
Why it matters: Claude's shift from chatbot to hands-on computer operator gives Pro and Max subscribers genuine task automation, but Anthropic itself disables sensitive-data apps by default and warns of errors — a hedge that signals even the builder doesn't fully trust the system. Security researchers note agentic AI can execute dramatic actions with little warning, and Anthropic's prompt-injection safeguards are still untested in the wild.
