Claude takes desktop control on macOS, rivaling OpenClaw

Anthropic says its Claude assistant can now control a desktop on macOS: it can click, type, open apps, and browse the web to complete tasks. The feature includes searching for and sending local files, filling spreadsheets, and navigating websites. Anthropic positions it as a simpler, “out-of-the-box” alternative to the open-source OpenClaw project, which gained traction via messaging-based instructions but is harder to set up (API keys, terminal work). For safety, Anthropic draws a clear line against OpenClaw’s broad system access. Claude requires a one-click enablement and asks for authorization per app rather than granting full machine access. Anthropic also claims built-in defenses against prompt-injection attacks, and it warns users not to let Claude access apps holding sensitive data for now; Claude should ask before opening new applications. Business customers using Claude Code can set tool-allow rules. Anthropic also introduced “Dispatch” in Claude Cowork: users can assign a task from a phone, step away, and return to finished work on their desktop. Market context: the agent race has already moved equities. Earlier coverage tied AI tool “computer control” to pressure on some Indian IT stocks. In China, OpenClaw reportedly spread via self-run Mac Mini setups, prompting later guidance for state-owned companies to stop using it on office machines. Claude is rolling out as a research preview on macOS for Claude Pro and Claude Max users, reflecting the ongoing shift toward AI agents—where trader-relevant risk comes mainly from how these tools change enterprise workflows and sentiment around AI automation.