Claude can now move your mouse. That’s a different kind of AI capability than generating text.
Anthropic launched Claude Computer Use as a research preview on or around March 24–25, 2026. The feature is available to Claude Pro and Max subscribers on macOS. Per coverage from the Daily AI News newsletter, Anthropic has introduced Computer Use as a capability enabling autonomous task completion across a user’s system, positioning Claude as what the newsletter describes as a meta-agent for AI automation.
What Claude Computer Use Actually Does
According to coverage from Engadget and PCMag, Claude can directly control a user’s browser, mouse, keyboard, and screen to complete tasks. The capability reportedly includes related features described in reporting as Cowork and Code, though the exact boundaries of what’s bundled under the Computer Use umbrella should be confirmed against Anthropic’s official announcement, which had not been located at time of publication.
The research preview framing is specific and deliberate. “Research preview” means Anthropic is gathering real-world data on how the feature behaves before moving toward general availability. For enterprise deployment decisions, that status matters: this is not a production-ready automation tool. It’s an early-access capability that carries the testing caveats that come with any research-phase product.
Why OS-Level Access Is the News
Most AI tools operate at the application layer, they read inputs and produce outputs within a defined interface. Computer Use operates at a different level. Claude can observe your screen, take control of applications you have open, and execute multi-step tasks across your system. That’s a meaningful capability expansion.
The practical question for enterprise teams isn’t whether this is impressive. It is. The question is what authorization framework applies when an AI model has direct OS access. What tasks should it be permitted to execute autonomously? What triggers a human-in-the-loop checkpoint? Those questions don’t have standard answers yet, and they’re the right ones to be asking before any enterprise pilot.
What “Research Preview on macOS” Means in Practice
The initial availability is macOS-only, per Engadget’s reporting. Claude Pro and Max subscribers can access it within their existing subscription, no additional cost was reported. The feature is available now, not on a future roadmap.
One newsletter analyst described the launch as “a major leap in practical, agentic AI deployment”, though that characterization reflects the author’s editorial view, not Anthropic’s own framing. The confirmed facts are more specific and more useful: OS-level task automation, research preview status, macOS availability, Pro and Max subscribers only.
What to Watch
The first thing to watch is Anthropic’s own documentation. An official technical announcement from Anthropic would clarify capability boundaries, security model, and known limitations, all of which matter for anyone evaluating whether to test this in a work environment.
The second is the competitive response. Agentic AI at the OS layer is a direction multiple companies are pursuing. Oracle embedded agents in the database this same week. The architectural question, where should agents operate, is actively being answered by the industry right now, with real products.
Enterprise teams considering a pilot should treat research preview status as a signal to test carefully, not to deploy broadly. The capability is real. The production governance model isn’t established yet.