On April 3, 2026, Anthropic unveiled an expansion of Claude’s desktop control feature to Windows, enhancing the AI assistant’s capabilities for Pro and Max subscribers using Claude Cowork and Claude Code. This development allows users to seamlessly operate applications, navigate web pages, and manage files directly on their PCs without any prior configuration. The feature, which is currently in research preview, also introduces a new Dispatch companion that enables users to assign tasks to Claude from their mobile devices, returning to find the work completed on their desktops without any manual intervention.
How Claude’s Desktop Control Works
Claude employs a structured tool hierarchy when executing tasks, prioritizing connectors like Slack and Google Calendar for structured tasks. Only when no dedicated interface exists does it resort to direct desktop manipulation. This tiered approach ensures that desktop automation is a fallback rather than a default action. When Claude does engage in direct control, it can manage a machine’s browser, mouse, keyboard, and screen, allowing it to perform tasks such as opening applications, navigating websites, and editing local files. Notably, no setup is required; the feature integrates with existing software without the need for API keys or configurations. Users must opt in to activate this feature, and Claude requests explicit permission before accessing new applications, with certain sensitive apps restricted by default for added security. However, Anthropic cautions that the feature may still harbor errors and advises against using it with applications that handle sensitive data, reflecting the company’s assessment that desktop automation is not yet fully reliable for critical workflows.
Initially launched as an API-only beta alongside Claude 3.5 Sonnet in October 2024, the desktop control feature has evolved into a consumer-facing product over the past 18 months. Following the introduction of Claude Cowork with file-editing capabilities in January 2026, the rollout continued with enterprise plugins in February, culminating in the macOS launch on March 24 and the rapid Windows expansion just ten days later. Each step has broadened access while maintaining the research preview label.
From Acquisition to Launch in Four Weeks
The technology behind desktop control is partly derived from Anthropic’s acquisition of Vercept AI, a startup specializing in AI-driven computer control. Vercept AI’s team, known for training models to interact with graphical user interfaces, shipped its first product within four weeks of joining Anthropic, dispelling concerns that corporate integration would hinder development. Co-founder Kiana Ehsani expressed her enthusiasm for the collaborative environment at Anthropic, noting the supportive culture that facilitated a smooth transition and rapid progress. This acquisition has proven strategic, as Vercept AI’s expertise in GUI interaction models is crucial for enhancing Anthropic’s desktop automation capabilities.
Security Questions Shadow the Expansion
Despite the swift rollout, the speed has raised security concerns. Shortly after the launch of Cowork, researchers demonstrated a file-stealing prompt injection attack, revealing vulnerabilities that could be exploited through malicious content in documents or web pages. With the extension of full desktop control to Windows, the potential attack surface expands significantly, encompassing not just file access but also mouse control, keyboard input, and screen capture. Anthropic has implemented safeguards, including prompt injection scanning and default restrictions on certain applications, but acknowledges that the feature remains prone to errors. Users can halt Claude’s operations at any time, yet the company’s documentation admits that it lacks the capability to disable the technology remotely once it has begun executing tasks. This admission raises critical questions about the implications of granting an AI assistant full desktop access, particularly in scenarios where intervention may be necessary.
As Anthropic navigates these challenges, it is not alone in the pursuit of desktop-level AI automation. Competitors like Microsoft and Google are also exploring similar capabilities, albeit with caution, keeping their implementations in limited preview stages. The industry continues to grapple with concerns surrounding data privacy, error rates, and controllability, and neither company has committed to a timeline for broader availability. The future of desktop automation hinges on proving its reliability and security for real-world productivity, a determination that will shape Anthropic’s path forward in this evolving landscape.
