Anthropic has launched computer use for Claude, giving the AI model the ability to directly control a user's mouse, keyboard, browser, and screen to complete tasks autonomously. The feature is now available in research preview for Claude Pro and Max subscribers through Claude Cowork and Claude Code, and is currently supported on macOS only.

For developers and knowledge workers who rely on Claude daily, this changes what they can realistically delegate. Rather than waiting for a dedicated connector to every app they use, Claude can now navigate software directly on screen: opening files, interacting with browsers, running development tools, and completing multi-step workflows without requiring manual handoff at each step. The launch also coincides with an update to Dispatch, Anthropic's mobile-to-desktop task delegation tool, which now works in both Cowork and Code.

In short, Anthropic has moved Claude from a tool that generates and advises into one that can act, navigating real desktop environments the same way a person would, on behalf of the user.

Computer use refers to an AI model's ability to perceive and interact with a graphical user interface, including a browser, file system, and applications, by simulating mouse movements, keyboard input, and screen navigation.

Here is what Anthropic announced, how the computer use capability works, and what it means for Claude Pro and Max subscribers using Cowork and Code.

Anthropic has introduced computer use as a new capability within Claude Cowork and Claude Code, giving Claude the ability to directly control a user's desktop environment. Available now in research preview for Claude Pro and Max subscribers, the feature allows Claude to interact with a user's browser, mouse, keyboard, and screen to complete tasks on their behalf, with no setup required.

When a user assigns a task, Claude will reach for the most precise tool first, starting with connectors to services like Slack or Google Calendar. If no connector exists for the required app or platform, Claude takes over the screen directly, reading what's on the display and interacting with it the same way a person would: scrolling, clicking, and opening files as needed. The company emphasized that Claude will always request explicit permission from the user before accessing new applications, and users retain the ability to stop Claude at any point during a task.

For now, the feature is limited to macOS and requires the Claude desktop app to be awake and running. Users must enable computer use through the desktop app settings before the capability becomes active.

Anthropic built safeguards into the system specifically designed to address the risk of prompt injection, a type of attack in which malicious instructions embedded in on-screen content attempt to redirect the model's behavior. According to Anthropic, the system automatically monitors Claude's internal processes during computer use sessions to detect any such attempt to redirect its behavior.

Despite these protections, Anthropic was direct in noting that computer use remains early-stage compared to Claude's existing text and coding capabilities. The company acknowledged that Claude can make mistakes, that complex tasks sometimes need a second try, and that working through a screen is inherently slower than using a direct integration. Anthropic also noted that while its safeguards continue to improve, threats are constantly evolving. As a precaution, some applications are off-limits by default, and Anthropic recommends users begin with trusted apps and avoid working with sensitive data while the feature is in research preview. Anthropic also directs users to its safety best practices documentation for further guidance.

Introduced last week, Dispatch is a feature available in both Claude Cowork and Claude Code that enables a continuous conversation between a user and Claude across devices. Users can assign a task to Claude from their phone, step away, and return to completed work on their desktop, without needing to re-establish context.

With the addition of computer use, Dispatch becomes a more powerful delegation tool. Anthropic described several practical scenarios that computer use and Dispatch make possible together: creating a morning briefing while a user is commuting, making changes in an integrated development environment (IDE), running tests and opening a pull request, or advancing a project like a 3D printing workflow according to a user's previously established plan.

Anthropic also noted that Dispatch can be configured for recurring tasks, such as automatically checking emails each morning or pulling metrics on a weekly basis. Users can also use Dispatch to spin up a Claude Cowork or Claude Code session for longer-form work like reports or pull requests.

The computer use capability is directed at Claude Pro and Max subscribers who work within desktop environments and need Claude to take action, not just provide guidance, across tools that do not yet have dedicated connectors. For developers, the ability to have Claude navigate an IDE, run tests, and submit a pull request without manual handoff removes friction from multi-step workflows. For knowledge workers, the ability to delegate recurring tasks like morning email reviews or weekly reporting through a mobile device represents a meaningful move toward ambient task execution.

Anthropic's decision to launch in research preview rather than as a general release reflects a deliberate approach to gathering real-world feedback. The company was direct in calling this a learning exercise, stating it is sharing the feature early to understand where it works and where it falls short, the same methodology applied during the Claude Cowork launch.

The macOS-only restriction and the requirement that the desktop app remain active and running are practical constraints that currently limit the feature's reach. Anthropic did not announce a timeline for expanding computer use to other operating systems or for moving the feature out of research preview.

Q: What did Anthropic announce about Claude's ability to interact with computers? A: Anthropic announced that Claude can now directly control a user's browser, mouse, keyboard, and screen to complete tasks within Claude Cowork and Claude Code. The feature is available in research preview for Claude Pro and Max subscribers on macOS and must be enabled through the desktop app settings.

Q: How does Claude decide when to use computer control versus a direct integration?
A: Claude prioritizes the most precise tool available first, starting with connectors to specific services like Slack or Google Calendar. When no connector exists for the required application or task, Claude moves to direct screen interaction, navigating the interface visually, clicking, scrolling, and opening files as a user would.

Q: Why is Anthropic launching computer use now, and why does it matter for users today?
A: The launch reflects growing industry expectations that AI models will not just respond but take action within the tools users already work in. For Claude Pro and Max subscribers, computer use closes the gap between what Claude can recommend and what it can actually execute, particularly in workflows that span multiple apps without dedicated connectors. Anthropic paired the launch with Dispatch, compounding the practical value by enabling mobile task assignment with desktop execution.

Q: What are the current limitations of Claude's computer use capability?
A: Computer use is currently supported on macOS only and requires the Claude desktop app to be running and enabled through settings. Anthropic acknowledged the capability is early-stage compared to Claude's text and coding performance — complex tasks may need a second attempt, and screen-based interaction is slower than direct integrations. Some applications are off-limits by default, and Anthropic has not announced a timeline for expanding to other operating systems or moving out of research preview.

Q: What safeguards does Anthropic have in place to protect users during computer use sessions?
A: Anthropic built protections specifically targeting prompt injection, a type of attack where malicious instructions embedded in on-screen content attempt to redirect the model. The system automatically scans model activations during computer use to detect this activity. Users must grant explicit permission before Claude accesses new applications and can stop Claude at any point during a task.

Q: How does Dispatch work alongside the computer use capability?
A: Dispatch enables a continuous conversation between a user and Claude across their phone and desktop, allowing tasks to be assigned from a mobile device and completed via computer use while the user is away. Anthropic described use cases including mobile-assigned morning briefings, IDE-based development tasks with automated pull requests, and recurring automated workflows such as daily email checks or weekly metric pulls.

Anthropic's computer use launch moves Claude from a tool that advises and generates into one that acts directly and autonomously within the apps and interfaces users already work in.

Key point: Claude can now complete tasks in desktop environments without requiring a custom integration for every application. This closes a meaningful gap between what an AI assistant can recommend and what it can actually execute on a user's behalf.

Who should care: Claude Pro and Max subscribers who work across multiple desktop tools, particularly developers using Claude Code and knowledge workers using Claude Cowork, will feel this most directly. The practical benefit is greatest for anyone who regularly hands off repetitive or multi-step workflows, since computer use and Dispatch together allow Claude to handle those tasks from start to finish without requiring the user to stay involved.

Why this matters now: This launch coincides with growing industry momentum around agentic AI, where the expectation is increasingly that models will not just respond but act. Anthropic is entering that space with a measured, safeguard-focused approach.

What decision this affects: For teams and individual users evaluating which AI tools to build workflows around, computer use changes the decision entirely. A tool that can operate across an existing desktop environment without requiring every app to have a dedicated integration reduces the barrier to adoption and expands the range of tasks that can realistically be delegated.

In short, Anthropic's computer use capability is not a feature addition. It is a category expansion. Claude is no longer limited to what it can say or generate; it can now navigate, click, and act within the software environments where real work happens. Combined with Dispatch's mobile-to-desktop task delegation, the two features together point toward a model of AI assistance that operates continuously, not just on demand.

The question is no longer whether AI can help with a task. The question is whether AI can take over the task entirely.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from Claude, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to Claude for assistance with research and editorial support in crafting this article.