Anthropic previews Model Hardware Standard

MHS could become critical infrastructure for physical AI, but its success depends on manufacturers adopting common drivers and robust safety controls.

• –Standardized read/write primitives make instruments discoverable and controllable across networks
• –MCP, CLI, and APIs let agents coordinate microscopes, liquid handlers, robotic arms, and lasers
• –Early partners include Genentech, HHMI Janelia, QuEra, AWS, Tecan, and Universal Robots
• –Reported tests include improving QuEra laser stabilization from 58% to 99.3%
• –Anthropic is gating the preview to develop safety evaluations before open-sourcing the standard