Anthropic Opens a Research Preview of the Model Hardware Standard (MHS): A Shared Specification for AI Agents to Safely Operate Physical Devices
What changed
Anthropic has opened a research preview of the Model Hardware Standard, a shared specification that helps AI agents find and operate physical devices safely. The standard puts safety limits in the device driver and reports results including an eight-hour instrument setup at Carnegie Mellon and laser relocking that improved from 58% to 99.3% across 700 trials at QuEra.
What this means for you
Researchers and organizations working with laboratory equipment may eventually connect devices to AI agents with less custom integration work. It is currently a research preview, so access and real-world availability may be limited.
