On August 27, Anthropic released the Model Hardware Standard (MHS), marking its first public entry into the fields of embodied intelligence and physical AI. The standard is a unified set of control and communication protocols that enable AI agents to operate devices equipped with programmable interfaces. Through this "translation layer," agents can simultaneously control microscopes, liquid handlers, robotic arms, and even quantum computing devices with minimal human intervention. In early closed-door tests, MHS demonstrated significant speed improvements in fields such as biopharmaceuticals, quantum computing, and advanced manufacturing. Genentech, a leading pharmaceutical company, reported that enabling AI to directly control multiple heterogeneous experimental devices reduced the development and execution cycle of fully automated dose-response curve experiments to one-third of the original time.

Author and source: AIBase

On August 27, Anthropic released a blog post introducing the "Model Hardware Standard" (MHS) in research preview form. Media outlets such as Reuters and CNBC interpreted this as the company’s first public foray into embodied intelligence and physical AI—AI systems capable of perceiving the physical world, executing complex actions, and continuously learning, applicable to domains ranging from autonomous driving to robotic arms.

Unified Translation Layer for Giving AI a "Body"

Similar to the earlier Model Context Protocol (MCP), MHS is essentially a unified set of control and communication standards that enable AI agents to operate devices with programmable interfaces. With this “translation layer,” agents can simultaneously control microscopes, liquid handlers, robotic arms, and even quantum computing devices, running workflows autonomously around the clock with minimal human intervention.

During early closed-door testing, MHS demonstrated exceptional acceleration capabilities in hard tech fields such as biopharmaceuticals, quantum computing, and advanced manufacturing. Genentech, a pharmaceutical giant, reported that enabling AI to directly control multiple heterogeneous lab devices reduced the development and operation cycle of fully automated dose-response curve experiments to one-third of the original time. Anthropic’s step of moving AI from the digital realm to physical production lines also signifies that the competition among large models is expanding from “conversation” to “action.”