Anthropic has launched a research preview of the Model Hardware Standard (MHS), a shared specification designed to allow AI agents to discover and safely operate physical devices. The initiative targets the "integration tax" caused by disparate vendor interfaces, aiming to reduce setup times from weeks or months down to hours or minutes.

  • MHS standardizes the driver layer between an OS and a device, exposing primitives like read, write, and discovery.
  • It uses driver tags to encode physical knowledge, such as weight limits, into reference files for safety enforcement.
  • Genentech used Claude to automate a BCA protein assay, converging on flow parameters confirmed by experts.
  • QuEra Computing replaced a bespoke laser-relock script with an MHS agent loop that achieved 99.3% success and reduced residual error from 15.7 mV to 1.55 mV.
  • Carnegie Mellon completed dose-response experiments in eight hours, roughly three times faster than typical vendor setups.

The standard is model-agnostic and compatible with the Model Context Protocol, allowing any agent harness to interact with devices via CLI or code files while keeping safety limits within the driver rather than the prompt.