Research preview opens for agent-to-machine control

Anthropic has opened the first phase of a research preview for the Model Hardware Standard, a new interface for AI agents to safely operate physical equipment in scientific research and advanced manufacturing. The preview was announced in a post on Twitter.

At a glance

• The standard defines how an agent communicates with laboratory instruments and production machinery, and according to CNBC it targets any device that exposes a programmable interface rather than one vendor's hardware line.
• Elizabeth Kelly, head of beneficial deployments at Anthropic, told CNBC the standard was built for science to demonstrate the promise of AI, with additional benefits for enterprise and industry.
• Access is limited to the research preview for now, though Anthropic intends to open source the specification later, according to CNBC, as it previously did with the Model Context Protocol.

Agent tooling has so far stayed inside software: files, interfaces, browsers. Extending it to instruments and production lines shifts the failure mode from a wrong answer to a wrong physical action, which is likely why Anthropic is presenting MHS as a safety standard rather than a capability release. The staged preview also reads as an attempt to fix the connector layer for lab automation and industrial equipment before a competing specification takes hold.

Anthropic describes MHS as a standard for AI agents to safely operate physical equipment, and states that the current release is the first phase of a research preview rather than a general launch. The scope named in the announcement covers scientific research and advanced manufacturing.

According to CNBC, MHS is designed to work with any device that has a programmable interface, a scope that takes in research instruments and equipment on factory floors alike. CNBC also reports that Anthropic plans to open source the standard, which for the moment stays inside the preview.

Elizabeth Kelly, who leads beneficial deployments at Anthropic, told CNBC that the standard was built for science in order to show the promise of AI, and that enterprise and industrial users stand to gain from it as well.

The Model Context Protocol, Anthropic's earlier interface for linking assistants to data systems, went to the Agentic AI Foundation on December 9, 2025, after being released as an open specification. MHS covers the equivalent connection to physical machinery, and stays under Anthropic's control during the preview.

Timing of the open-source release

No date has been given for moving MHS out of the research preview, and CNBC frames the open sourcing as an intention rather than a scheduled step. The announcement does not list participating institutions or equipment makers, nor does it set out what the phases after the first one will add. Pricing and licensing terms for the eventual release are not stated either.