Anthropic just took its first real step outside of software. The company launched a research preview of what it calls the Model Hardware Standard, or MHS: a common language that lets an AI agent like Claude talk directly to physical machines such as robotic arms, microscopes, and lab liquid handlers, instead of just reading and writing text.
The problem MHS is trying to solve is simple to picture. Most scientific and factory equipment cannot talk to other equipment out of the box. Getting a laser, a camera, and a robotic arm to work together on one experiment usually means hiring a specialist to write custom connector software, and that process can drag on for weeks or months. MHS gives every device a shared plug, so an AI agent can read what a machine does and operate it without that custom bridge.
The idea reportedly came from watching a neuroscientist at the Janelia Research Campus spend enormous effort wiring together lasers, microscopes, and cameras for a single memory experiment. Anthropic saw that same wiring problem everywhere in science and manufacturing, and decided to build a fix once rather than let every lab solve it alone.
This is not Anthropic's first attempt at building a shared connector. In 2024 it released the Model Context Protocol, a standard that let AI assistants plug into software tools like email and calendars. That protocol was adopted so widely, by competitors included, that it became the default way AI agents connect to digital tools. MHS is the same bet, aimed at physical equipment instead of software.
The early test group is telling. Amazon Web Services, lab equipment maker Danaher, Hugging Face, and Raspberry Pi are already trying it, alongside drugmaker Genentech, where a scientist reportedly handed Claude a PDF describing an experiment and the AI ran it on real lab hardware without further help. That is a meaningful jump from AI that writes reports to AI that runs equipment overnight.
It also puts Anthropic on a path where OpenAI and Amazon are already spending heavily, building AI-linked devices and manufacturing tools of their own. Whoever sets the plug that every machine uses tends to keep a lasting advantage, the same way Windows shaped personal computers for decades.
For any business running a mix of machines, whether that is a lab, a factory floor, or a warehouse, the promise is fewer specialists needed to make old and new equipment work together, and faster setup for new production lines or experiments. That is real money saved on integration work that used to take months.
There is a catch to watch, especially for operations in Europe. Starting January 2027, new European machinery rules will treat AI-controlled safety functions as regulated safety parts. That means whoever writes the file telling a robotic arm how fast it can move may now carry legal responsibility if that file gets it wrong. Anyone adopting MHS for real equipment should plan for that paperwork now, not after the rule takes effect.