AI Just Got Scary 🤖🤯 Autonomous Control?
AI
🎧
Anthropic has recently updated its Claude AI tools, Code and Cowork, to operate autonomously using a user’s computer. This new feature, available as a research preview for Claude Pro and Max subscribers, allows the chatbot to automatically open files, utilize web browsers and applications, and run development tools. The capability builds upon previous autonomous features introduced in Claude’s 3.5 Sonnet AI model in 2024. Currently, the feature is limited to macOS devices and prioritizes connections to services like Slack and Google Workspace. The chatbot will seek explicit permission before actions and works particularly well with Dispatch, a cross-device task assignment tool. Anthropic is sharing this update early to gather data on its performance, acknowledging that complex tasks may require a second attempt due to its slower, screen-based operation.
AUTONOMOUS COMPUTER CONTROL IN CLAUDE
Anthropic has significantly enhanced Claude’s capabilities by enabling it to autonomously control a user’s computer, accessible through its Code and Cowork AI tools. This new feature, currently in research preview for Claude Pro and Max subscribers, allows Claude to perform tasks such as opening files, navigating web browsers, utilizing applications, and running development tools – all without requiring any prior setup. This functionality expands beyond the chatbot interface, offering a truly integrated workflow for developers and users alike.
KEY FUNCTIONALITY AND ACCESS
The core of this autonomous control lies in Claude’s ability to interact with the user’s computer directly. Specifically, the feature leverages connectors to services like Slack and Google Workspace, prioritizing these integrations for efficiency. However, if a connector isn’t available, Claude will utilize direct control over the user’s browser, mouse, keyboard, and display. This ensures that Claude can execute tasks regardless of the availability of pre-built integrations. Access to this feature necessitates the running of the Claude desktop app on a supported macOS device, paired with the Claude mobile app. Anthropic emphasizes that the current implementation is limited to macOS, reflecting a deliberate approach to research and development.
USER PERMISSION AND LIMITATIONS
A crucial element of the design is Claude’s commitment to user safety and control. The system is programmed to consistently seek explicit permission before undertaking any actions, including browsing, scrolling, and clicking. This proactive approach minimizes the risk of unintended consequences and maintains user agency. While the feature performs exceptionally well with Dispatch, Anthropic acknowledges that complex tasks may occasionally require a secondary attempt, and that direct screen interaction is inherently slower than utilizing a dedicated integration. The company is actively soliciting feedback to refine the system and identify areas for improvement, prioritizing a phased rollout of this innovative technology.
This article is AI-synthesized from public sources and may not reflect original reporting.