Anthropic opens a Model Hardware Standard research preview for lab and factory devices

Anthropic opened a gated research preview of the Model Hardware Standard (MHS), a shared specification it says will let AI agents operate programmable lab and factory devices.

A shared spec for agents that drive microscopes and robot arms is how software starts touching wet labs and factory floors. The confirmed event is the gated preview; partner performance numbers stay theirs.

Anthropic published “Previewing the Model Hardware Standard,” opening a research preview of MHS to a first group of scientific research labs and advanced manufacturers. That is the announcement. Access is application-gated via Anthropic’s page and modelhardwarestandard.com.

Anthropic describes MHS as a shared specification so AI agents can safely operate physical devices — the page names microscopes, liquid handlers, and robotic arms. Development began as a collaboration between Anthropic and HHMI (Howard Hughes Medical Institute) Janelia Research Campus.

The company says MHS works with devices that have a programmable interface, is model-agnostic — not tied to one lab’s model — and can be reached via protocols such as MCP (Model Context Protocol), a common way for models to call external tools. Those are Anthropic’s product claims.

Early partner write-ups on the same page include Genentech, the University of Washington Baker and Pinglay labs, HHMI Janelia, QuEra, and others. Partner performance numbers — including QuEra’s laser-lock recovery rate — stay their reports on Anthropic’s page.

Anthropic says it will open-source the standard after preview partners help build safety evaluations. No public spec or repo is generally available yet, and the page does not state an open-source date.