Anthropic previews a standard for AI-controlled lab and factory hardware

Anthropic has opened a research preview of the Model Hardware Standard (MHS), a shared specification for AI agents operating physical equipment. As Anthropic News reports, the first preview is available to selected scientific research labs and advanced manufacturers ahead of a planned open-source release.

MHS is designed to replace weeks or months of bespoke hardware integration with setup that can take hours or minutes. It supports programmable devices including microscopes, liquid handlers, robotic arms, and other laboratory or manufacturing equipment, allowing agents to coordinate several machines at once.

A common interface for physical systems

The standard uses software drivers with basic “read” and “write” operations, while making devices discoverable in a consistent format. Natural-language tags can describe details that code often misses—such as a robot arm’s weight, adjustable settings, and safety limits—and generate a reference file an agent can use.

Agents can control connected equipment through the Model Context Protocol (MCP), a command-line interface, or code files. That combination lets an agent monitor results, change parameters during an experiment, and hand repeatable routines to deterministic scripts. Anthropic says Claude, for example, explored laser alignment through repeated adjustments and camera observations before packaging the learned process into a one-command script.

Early projects included a Genentech proof of concept coordinating a liquid handler, robotic arm, and plate reader for a protein assay. Researchers at Carnegie Mellon reported running serial-dilution dose-response experiments about three times faster, while University of Washington teams used MHS for remote monitoring, AI-supervised qPCR, and safer plate transfers.

For AI builders, the preview points toward a model-agnostic layer between agents and real-world machines. It also highlights the need for safety evaluations and operating practices before such systems are broadly deployed.

Source: Anthropic News