Anthropic tests a new standard for Claude to work with factory and lab hardware

Anthropic has released the Model Hardware Standard in research preview, a specification that stores how a piece of hardware works and how an AI may safely operate it, such as limiting a robot arm’s speed and angles. Europe’s Machinery Regulation starts covering AI-based safety functions on 20 January 2027.

Anthropic has written a standard for telling an AI how a machine works. The Model Hardware Standard, out in research preview, stores what a piece of hardware does and how it may be operated, so a model does not have to rely on a paper manual or on the two people who know.

The example the company gives is a robot arm. A vendor can specify in an MHS file how an AI may move a heavy arm safely, limiting the speeds and angles it is allowed to use.

The pitch is about knowledge that exists but cannot be reached. “In many cases, the science doesn’t happen because you can’t use the equipment,” said Jonah Cool of Anthropic’s life sciences arm.

The precedent is Anthropic’s own. “What MCP did for software, MHS will do for the hardware world,” said Alek Kemeny, referring to the Model Context Protocol that now connects assistants to Gmail, Calendar and Slack.

The demonstration is a laboratory. Anthropic says a Genentech scientist sent Claude a PDF of an experiment they had designed, and the assistant executed it autonomously on hardware carrying an MHS specification.

The test group is substantial. Amazon Web Services, Danaher, Hugging Face and Raspberry Pi have been trying it, and a waitlist is now open to anyone else.

Two of those are European in origin. Raspberry Pi is British and listed in London, and Hugging Face was founded by a French team.

The machines are increasingly European too. NEURA Robotics raised up to $1.4B in Germany this year, in a humanoid sector TNW has argued is quietly better placed than it looks.

Europe is about to regulate this precise thing. Regulation 2023/1230 replaces the Machinery Directive entirely on 20 January 2027 and covers AI-based safety functions and machinery with self-evolving behaviour for the first time.

That changes what an MHS file might be. A specification that constrains a robot arm’s speed and angles is performing a safety function, and for high-risk categories self-declaration of conformity stops being enough.

The safety record is worth holding in mind. TNW reported that Claude Cowork could escape its virtual machine and read credentials on a Mac, and that was software.

Anthropic intends to open-source the framework once the preview ends. In Europe, whoever writes that file may find they have written a regulated safety component.