Anthropic's Model Hardware Standard

Anthropic launched a research preview of the Model Hardware Standard (MHS) on August 27, 2026, targeting scientific research labs and advanced manufacturers. MHS provides a common interface with drivers exposing read and write commands, plus reference files describing device capabilities, physical properties, and enforced safety limits. AI agents access MHS drivers via MCP, command-line tools, or code, and can orchestrate multiple instruments in parallel, sequence experimental steps, monitor results, and in some cases recover from hardware errors without human intervention.

The standard is model-agnostic, meaning users are not locked into Claude models. Anthropic describes MHS as analogous to USB-C, standardizing how information is transmitted between devices. The concept originated from observing neuroscientist Arco Bast coordinate lasers, microscopes, and cameras for a brain memory-formation experiment at HHMI Janelia Research Campus, where MHS was developed in collaboration.

Hardware vendors Tecan, Universal Robots, and Doosan Robotics are adding native MHS support; launch partners include AWS, Hugging Face LeRobot, QIAGEN, and Raspberry Pi. Access during the research preview requires joining a waitlist at modelhardwarestandard.com.

Early results reported by Anthropic include automated protein assays, dose-response experiments running three times faster, and laser stabilization at QuEra Computing where an agent-built script recovered a quantum laser's lock in 695 of 700 trials without human intervention. Anthropic also reports observing Claude make iterative adjustments to a laser, using a camera to assess results and repeating the process.

MHS reduces hardware integration time from weeks or months to hours or minutes, Anthropic reports. MHS specifications can restrict an agent from using particular hardware, functioning as a safety mechanism. Anthropic notes that Claude's spatial and physical reasoning have limitations that still require expert oversight. Physical hardware carries different stakes than software; bad commands can cause physical damage rather than just software errors.

Anthropicplans to open-source MHS after the research preview, using the interim period to develop safety evaluations, a physical safety roadmap, and best practices. The standard is designed to expand beyond lab and manufacturing equipment to boards, cameras, and hobbyist hardware. Anthropic previously open-sourced the Model Context Protocol in 2024.