- Anthropic launches AI that controls computers through clicks and keystrokes
- Early tests show promise but reveal significant limitations
- Safety measures protect against misuse but challenges remain
Your Computer Has a New Co-pilot
Claude 3.5 Sonnet, Anthropic’s latest AI model, can now navigate desktop applications through keystrokes, clicks, and mouse movements.
The upgraded version arrives through a Computer Use API beta, marking a significant leap in AI automation capabilities.
Not Just Another Automation Tool
Despite entering a crowded market of AI agents, Anthropic’s approach stands out. The model learns by analyzing screen pixels to determine cursor movements and can handle complex, multi-step tasks.
Early adopters include Replit, which created an autonomous app verifier, and Canva, exploring AI-powered design support.
Safety First, But Risks Remain
Anthropic acknowledges potential dangers and has implemented safeguards against misuse. The model includes classifiers to avoid high-risk actions and retains screenshots for 30 days.
Yet performance tests show room for improvement – the AI still stumbles on basic tasks like airline bookings and returns processing.