Anthropic launches Model Hardware Standard, cutting device integration time to hours

Anthropic's new Model Hardware Standard (MHS) tackles a practical bottleneck in lab and factory automation: devices from different manufacturers speak different protocols, making integration a slow, custom job. By standardizing how AI agents read from and control physical hardware, MHS aims to turn a weeks-long engineering task into an hours-long setup. The approach directly extends Anthropic's Model Context Protocol (MCP) for software, applying the same principle of a common interface to the physical world.

The partner tests show promise but also reveal the limits of current AI. At Genentech, Claude coordinated lab equipment to automate a protein assay but struggled when bubbles formed in a viscous solution, repeatedly restarting the process with tweaked parameters until a human explained the physical cause. This underscores Anthropic's own admission that Claude's physical reasoning is limited and still requires expert oversight. The company plans to build additional safety evaluations with partners during the research preview and is working on a physical safety roadmap to guard against misuse, highlighting that the path to fully autonomous physical AI is still a work in progress.