Agentic AI has so far mostly acted on screens; Anthropic's new effort aims to push those systems into the physical world by standardising how models talk to machines.
The company released a research preview of the Model Hardware Standard, or MHS, a specification of drivers, metadata tags and interfaces that present devices to models in a common format.
MHS is designed to remove bespoke integrations that currently glue together microscopes, lasers, cameras, robotic arms and other equipment, allowing instruments to "talk to each other across a network" without a custom translator.
Anthropic says that can cut painstaking integration work from weeks or months to hours or minutes.
The standard includes a standardized tagging system that encodes a device's physical limits and adjustable parameters, weight, reach, measurement modes and safety bounds, so a model unfamiliar with a piece of hardware can still reason safely about what it can and cannot do.
Anthropic also showed demos of a model sequencing commands, focusing a microscope, analysing an image, moving to a target, and getting a robotic arm to pick up an aluminium can after reasoning through the necessary steps, and says models can automate sequences by writing API scripts and modifying them as conditions change.
MHS can be used directly through command-line and API code, or connected to models via Anthropic's Model Context Protocol so AI systems can interact in natural language and update parameters in real time.
Anthropic positions the work as model-agnostic and eventually open: the preview involves a first group of labs and manufacturers to test safety and practice-building, with partners including Amazon Web Services, Hugging Face, Raspberry Pi, Automata and Universal Robots helping to build safety evaluations.
For researchers, the pitch is straightforward: faster iteration means faster experiments. Anthropic says early internal tests with scientific partners already showed quicker integration and faster cycles. The company frames that speed as transformative: in a video accompanying the announcement, Anthropic technical staffer Alek Kemeny said, "This is how a century of progress can condense into a decade."
Who should watch: labs and advanced manufacturers that depend on delicate, multi-part hardware; the preview period and Anthropic's safety evaluations will determine whether MHS becomes a practical, widely adopted bridge between models and machines.
