Anthropic reveals its view of how AI agents should interact with the physical world

Anthropic has introduced a preview of its Model Hardware Standard (MHS), a system designed to let AI agents control programmable physical machines like microscopes, robotic arms, and liquid handlers through shared software rules. MHS uses standardized software drivers that allow different laboratory and industrial equipment to communicate without needing custom software links, reducing integration time from months to just hours or minutes. During early trials, the Claude AI model successfully performed tasks such as laser alignment and coordinating multiple instruments. The standard has been shared with industry participants like AWS, Doosan Robotics, and Universal Robots to establish safety checks and practical deployment rules before a broader rollout. Despite its capabilities, Anthropic emphasizes that AI agents still require expert human supervision due to physical reasoning limitations. The project remains experimental, with plans for an eventual open-source release featuring expanded compatibility and detailed safety guidance.