Anthropic's Claude gains computer control in Code and Cowork tools
Anthropic has expanded Claude's autonomous capabilities to its Code and Cowork AI tools, allowing the model to control your Mac's mouse, keyboard, and display to complete tasks without manual intervention. The research preview is available now for Claude Pro and Max subscribers on macOS only, with support for other operating systems coming later.
Anthropic has updated Claude's Code and Cowork AI agents with autonomous computer control capabilities, allowing the model to manipulate your Mac's interface directly when needed.
The feature is now available as a research preview for Claude Pro and Max subscribers, though macOS is currently the only supported operating system. Users must have the Claude desktop app running on a compatible Mac and paired with Claude's mobile app to access the functionality.
How It Works
Claude can now autonomously open files, launch applications, control web browsers, and execute developer tools without requiring manual setup or input. The system prioritizes direct integrations with services like Slack and Google Workspace when available, but will fall back to directly controlling your mouse, keyboard, and display if a connector isn't accessible.
According to Anthropic, the model will request explicit user permission before performing any autonomous actions. The feature works across scrolling, clicking, and navigating interfaces as needed to complete assigned tasks.
The implementation integrates with Anthropic's Dispatch feature—a cross-device tool released recently that lets users assign tasks to Claude from their mobile devices. This combination allows users to delegate work to Claude even when they're away from their computer.
Current Limitations
Anthropicbuildsonwork introduced in Claude 3.5 Sonnet in 2024, which first demonstrated autonomous computer interaction capabilities. However, the company acknowledges significant limitations in the current research preview.
"Complex tasks sometimes need a second try, and working through your screen is slower than using a direct integration," Anthropic stated. The company is deliberately releasing the feature early to identify where the technology succeeds and where it struggles.
The feature is limited to macOS for now, with the company planning to expand to other operating systems in the future.
What This Means
This marks a significant step toward Claude functioning as a true autonomous agent capable of handling real-world computer tasks without human intervention. The research preview approach allows Anthropic to gather real-world usage data and improve reliability before wider rollout. However, the acknowledged performance gaps—particularly the need for task retries and slower execution compared to direct API integrations—suggest the technology remains in early stages. The macOS-only limitation also restricts immediate utility for developers and teams using Windows or Linux systems.
Related Articles
Anthropic Adds Cross-Session Messaging to Claude Code v2.1.224
Claude Code v2.1.224 introduces cross-session messaging, letting separate Claude Code instances on macOS and Linux send each other summaries to coordinate work. The feature does not support approving permissions or executing commands remotely.
Anthropic Cuts False Positives in Fable 5's Biology Filter by 85%, Keeps Virology and Toxicology Blocked
Anthropic has cut false positives in Fable 5's biology safety classifier by roughly 85%, letting users ask about lab results, symptoms, and medical questions without being rerouted to the weaker Opus 5 model. Dual-use topics like virology, toxicology, and molecular design remain restricted, with Anthropic citing the difficulty of containing biological threats once released.
Anthropic SDK v0.121.0 Adds Session Budgets, Mid-Conversation Tool Changes, and GitHub Skills Auto-Loading
Anthropic released version 0.121.0 of its Python SDK on August 7, 2026, introducing a new beta for mid-conversation tool changes, session budgets, an advisor tool, pinned inference location, and skills auto-loading from GitHub. The update also removes retired Claude Opus 4.1 models from the API.
Study: Humans Approve 1 in 3 Malicious AI Coding Agent Commands in Browser Game Test
A browser-based game simulating Claude Code-style permission requests found that human reviewers approved roughly one in three malicious commands across more than 40,000 game sessions. The findings, alongside Anthropic's own telemetry showing 93% approval rates for permission prompts, highlight growing concerns about approval fatigue in agentic AI coding workflows.
Comments
Loading...