Anthropic’s Claude controls your Mac
Anthropic rolled out a preview that lets Claude autonomously control your Mac—running apps, manipulating files, and finishing multi‑app workflows—with the feature available to Claude Pro and Max users for now. The capability is Mac‑only at launch and comes with updated APIs/SDKs and safety guardrails, signalling a new OS‑level frontier for agentic AI. ( )
Anthropic’s public rollout and media demos ran across March 23–24, 2026, with CNET publishing an explainer on March 23 that framed the update as an expansion of Claude’s “agentic” capabilities. (cnet.com)) Anthropic showed the new desktop interactions inside two offerings — Claude Code and Claude Cowork — demonstrating task handoffs and background sessions tied to those product surfaces. (9to5mac.com)) When a direct connector to a service isn’t available, Claude falls back to emulating a human user by moving the cursor, clicking, typing and scrolling inside apps and browsers to complete steps. (cnet.com)) The open-source claude-code repository changelog documents recent remote-control engineering work, including an added optional name argument for /remote-control sessions and multiple bug fixes to remote-control session titles and startup behavior. (github.com)) Anthropic’s platform release notes show concurrent API changes this month, such as adding model capability fields to the Models API on March 18 and general availability of a 1M‑token context window for Claude Opus 4.6 and Sonnet 4.6 on March 13. (platform.claude.com)) Security researchers and reporters flagged risks tied to agentic desktop control, warning about expanded attack surface, prompt‑injection vectors, and the ability for autonomous actions to execute quickly without manual oversight. (appleinsider.com)) Anthropic also rolled out a related “control Cowork from your phone” research preview starting March 17, 2026, noting a staged roll‑out that begins with Max plans and follows to Pro plans over subsequent days. (support.claude.com))
