Claude Model Hardware Standard (MHS) Connects AI Agents to Lab Equipment.

Anthropic is moving Claude closer to the physical world. The company has begun the first phase of a research preview for Model Hardware Standard (MHS), an interface designed to let AI agents discover and safely operate scientific and manufacturing equipment.

The standard tackles a practical obstacle facing laboratories that want to use AI beyond analysis and documentation. Connecting an agent to a microscope, laser system, liquid handler, or other specialized device can currently require weeks of custom integration. Anthropic says MHS could reduce that work to hours or minutes while giving agents a consistent way to identify devices and operate them within defined safety limits.

Anthropic announced the preview on LinkedIn, describing it as an early step toward a standard that will eventually be open sourced. For now, participation is limited to research-preview stakeholders across science, robotics, and advanced manufacturing.

Laboratory hardware is fragmented. Different devices expose different controls, data formats, operating constraints, and error states. An AI agent may be capable of planning an experiment, but it cannot reliably execute that plan unless developers first translate its instructions into commands that each machine understands.

MHS is intended to provide a common layer between agents and equipment. According to Anthropic, the interface makes compatible devices discoverable and communicates the boundaries within which an agent can operate them. That could allow an agent to respond to changing experimental conditions without receiving unrestricted control over the underlying hardware.

The approach builds on work between Anthropic's beneficial deployments team and Howard Hughes Medical Institute's Janelia Research Campus. It also extends Anthropic's broader push into research workflows, following the launch of Claude Science, which focused on shortening the cycle between scientific questions, analysis, and results.

Early MHS tests produced three notable examples:

• At Genentech, an AI agent ran a drug-discovery experiment and handled errors in real time.
• At HHMI Janelia Research Campus, researchers compressed an imaging experiment that had taken weeks into a single day.
• At quantum computing company QuEra, the standard was used in work that improved laser stabilization from 58% to 99.3%.

Those results suggest that the immediate opportunity is not simply automating a fixed sequence of machine commands. It is enabling agents to monitor experiments, detect problems, and adjust execution while remaining inside constraints set by researchers and equipment operators.

Anthropic is not yet presenting MHS as a finished standard. The company acknowledged that large language models lack robust physical intuition because their understanding of the world is primarily learned from text and images. A flawed response in a chat window is one thing; an incorrect command sent to laboratory or industrial machinery can damage equipment, invalidate an experiment, or create a safety risk.

The research preview will therefore be used to develop additional safety evaluations and strengthen protections before a wider release. Anthropic has not disclosed a public launch date, repository, licensing model, or commercial pricing. It is inviting scientific and industrial stakeholders to help shape the specification before the planned open-source phase.

For smaller hardware companies and specialist manufacturers, a widely adopted standard could eventually reduce the need to create separate agent integrations for every customer or AI system. The open questions are whether equipment vendors will implement it, how consistently safety limits can be represented across very different machines, and what oversight will remain necessary when agents encounter unfamiliar physical conditions.

What is Claude Model Hardware Standard (MHS)?

Model Hardware Standard is a proposed interface from Anthropic that helps AI agents discover and operate physical equipment used in scientific research and advanced manufacturing. It is designed to shorten device integration while restricting agents to defined operating and safety limits.

Who can use MHS now?

MHS is in the first phase of a research preview. Anthropic is inviting stakeholders from science, robotics, manufacturing, and related industries to participate, but it has not announced general public availability.

Is Model Hardware Standard open source?

Not yet. Anthropic says it intends to open source MHS after using the research preview to conduct more safety evaluations and improve protections for AI operating in the physical world. No release date or licensing details have been provided.