Anthropic launches Model Hardware Standard to let AI agents control physical machines

Anthropic is giving AI agents a way to reach beyond screens and software and start operating machines in the physical world.

The Claude maker on Thursday unveiled the Model Hardware Standard (MHS), a new specification that lets AI agents communicate with and operate programmable hardware, from microscopes and liquid handlers to robotic arms and equipment used inside quantum computers.

Announcing the launch on X, Anthropic said, “Today, we’re kicking off the first phase of the research preview for Model Hardware Standard (MHS): a new standard for AI agents to safely operate physical equipment in scientific research and advanced manufacturing.”

The idea is simple but ambitious: give machines from different manufacturers a common interface that AI agents can discover, read, and control without engineers building a custom integration for every device.

Anthropic says connecting equipment in labs and manufacturing facilities can take weeks or months today. MHS could cut that work to hours or minutes.

“We built this for science to sort of show the promise of AI, but there’s also huge benefits here for enterprise and for industry,” Elizabeth Kelly, Anthropic’s head of beneficial deployments, told CNBC.

MHS is starting as a research preview with organizations across science, robotics, electronics and manufacturing. Anthropic plans to open-source the standard after working with early partners on safety evaluations and best practices.

The standard is model-agnostic, meaning companies aren’t locked into Claude. Any AI agent can potentially interact with MHS-enabled equipment using supported interfaces, including Anthropic’s Model Context Protocol, command-line tools and APIs.

That distinction matters. MCP, which Anthropic introduced in 2024 and later open-sourced, gives AI systems a standard way to connect with software tools and data. MHS tackles a different layer: the hardware itself.

Put another way, MCP connects AI agents to digital systems. MHS could help connect them to physical machines.

From chatbots to machines that act

MHS introduces a standardized software driver that translates between computers and physical devices. Hardware exposes common commands such as “read” and “write,” allowing an agent to retrieve information, change settings, and coordinate multiple pieces of equipment.

The driver can contain natural-language information about a machine’s physical characteristics and operating limits. An AI agent encountering unfamiliar equipment could learn what the machine measures, which settings it can change and which safety boundaries it must respect.

Anthropic has already put the concept through real experiments.

At Genentech, researchers connected Claude through MHS to a liquid handler, robotic arm, and plate reader for a protein assay. Claude ran experiments, analyzed the results, and adjusted liquid-transfer parameters. It recovered autonomously from some equipment failures, though researchers found the model still struggled with physical phenomena such as bubbles in liquids, exposing a significant limitation of current AI reasoning.

At Carnegie Mellon University, researchers used MHS to connect laboratory equipment spread across three computers with incompatible interfaces. The team says building the system took roughly eight hours, compared with several weeks for a typical vendor-built setup. An AI agent then ran a dose-response experiment, rejected an inadequate result, changed the concentration range and repeated the experiment without human intervention.

Researchers reported the experiments ran about three times faster than before.

Another project at the University of Washington connected six instruments in less than a week. Researchers used an AI agent to monitor experiments remotely and coordinate a robotic arm with a liquid handler for automated sample transfers.

These are proofs of concept, not evidence that autonomous laboratories are ready for broad deployment. Anthropic acknowledges that current models can stumble over physical, chemical, and biological constraints that humans grasp through experience. Long-running agents also bring compute costs.

Anthropic wants a standard others can build on

The larger bet goes beyond Claude.

MHS works with any device that exposes a programmable interface, and Anthropic wants manufacturers to adopt the specification directly. An open standard could give robotics companies, scientific labs and manufacturers a common way to make their equipment accessible to AI systems.

That could become increasingly significant as the AI industry shifts from agents that generate text and operate software to systems expected to take actions in factories, laboratories and other physical environments.

Anthropic has already shown how much influence an interface standard can gain. MCP grew from an Anthropic project into a widely adopted method for connecting AI agents with external tools and data.

MHS is an attempt to carry a similar idea across the boundary between software and machinery.

If it catches on, the next phase of the AI agent race may be less about what models can say and more about what machines they can safely make move.

Below is a video showing MHS in action.