Claude can now use your computer — Anthropic's Cowork research preview

Anthropic announced Cowork, a research preview that enables its Claude AI assistant to perform tasks directly on users' computers when granted access. The company says Cowork allows Claude to act on local files and interact with browser sessions while requesting permission before significant actions.

In practice, users can give Claude access to a chosen folder and let the agent organize files, synthesize documents, or complete multi-step browser workflows using existing connectors. Anthropic positioned Cowork as an extension of Claude Code aimed at non-technical users and noted the preview is available to Claude Max subscribers on macOS for now. The company emphasizes control mechanisms and user approvals for sensitive operations.

The timing of the launch follows the viral rise of OpenClaw, an open-source agent platform whose rapid adoption has accelerated interest and competition in agentic AI. Industry observers say the OpenClaw phenomenon — and the resulting public scrutiny — is prompting major AI firms to accelerate controlled, productized agent features.

Security researchers have warned about a surge in malicious activity tied to agent ecosystems, including fake installers and data‑stealing plugins targeting popular agent distributions. Those warnings underscore the need for isolation layers, signed distributions and tightened permission models as agents gain the ability to read and write local resources. Anthropic’s preview appears designed to balance utility with explicit consent and sandboxing, though risks remain in practice.

From a market perspective, Cowork could push rapid adoption among productivity tool buyers and alter subscription value propositions for AI platforms. Analysts expect regulatory attention and enterprise governance frameworks to evolve in parallel; adoption will likely hinge on demonstrable security controls, clear consent flows and measurable productivity gains for both individual users and IT organizations.