Previewing the Model Hardware Standard

What changed

Source summary

Anthropic opened a research preview of the Model Hardware Standard (MHS), a shared specification letting AI agents like Claude safely operate physical lab devices (microscopes, liquid handlers, robotic arms) through one common interface, cutting hardware integration time from weeks to hours. It currently works best with Claude, does not support hardware lacking a programming interface, and Anthropic states Claude's spatial/physical reasoning still has limitations requiring expert oversight; access requires an application.

Why it matters to your work

Practical interpretation

If your organisation runs a research or lab team (university, pharma, biotech), this could eventually let AI agents run physical lab equipment directly instead of needing bespoke integrations per device. It is an early research preview requiring an application, works best with Claude, and Anthropic is explicit that human experts still need to oversee any physical actions — not yet relevant to typical office-based work.

When this could help

Useful when a lab or manufacturer wants to automate device operation with an AI agent instead of writing bespoke integration code per instrument, and is willing to join a research preview with current expert oversight requirements.

The Model Hardware Standard (MHS) is a shared specification that lets an AI agent safely discover and control lab hardware through one common interface, cutting hardware integration time from weeks to as little as about eight hours for partners who've tried it.

How to get started

1. Describe each device in natural language using tags in the MHS driver — write them directly or have an agent interview you about the setup.

   Provider guidance ↗

2. The driver automatically generates a reference file covering the device's measurements, adjustable settings, and enforced safety limits.

   Provider guidance ↗

3. Connect your devices — agents can then discover them, read their status, and issue commands via MCP, a command-line interface, or code files.

   Provider guidance ↗

4. Apply for research preview access or join the waitlist.

   Provider guidance ↗

What to check

• Anthropic states Claude's spatial and physical reasoning still has limitations and requires expert oversight — e.g. one partner had to guide Claude to treat sample foaming as a physical problem rather than a software bug, since its default was to retry and create more bubbles. Source ↗
• Another partner (QuEra) reported Claude could not troubleshoot physical hardware faults because its understanding of the rig was programmatic, and it often paused to ask for confirmation before even slightly risky actions. Source ↗
• MHS does not yet support hardware without a programmable interface; Anthropic is working with manufacturers to add MHS drivers over time. Source ↗
• Continuous agent monitoring of hardware carries compute costs that need to be weighed against researcher time saved. Source ↗