AI GlossaryㅁTechnical words in the news
Model Hardware Standard (MHS)
A common specification that lets AI agents operate real-world equipment, like microscopes or robotic arms, regardless of manufacturer.
In plain words
The Model Hardware Standard is a common specification that lets AI control not just software on a screen, but physical equipment like microscopes, robotic arms, and liquid handlers. It's often compared to USB-C: in the past, every laptop had a different charging port and needed its own cable, but now one port connects to everything. Lab equipment similarly speaks different "languages," so before this standard existed, connecting a single piece of equipment to AI could take specialists weeks of work.
At the core of this standard is a common translator that lets any piece of equipment call basic operations, like reading or setting a temperature, in the same way. Information such as a device's weight or its safe operating range is also attached as human-readable descriptions, so that even an AI encountering the equipment for the first time can operate it just by reading those descriptions.
Equipment connected this way can be controlled by an AI making its own decisions, operated directly by a person through a command line, or bundled into a single executable for repetitive tasks so they can be run quickly without having to be re-reasoned each time.
How it shows up in the news
On August 27, 2026, Anthropic released a research preview of this standard, showcasing real-world cases where it was tested, such as automating protein analysis at Genentech and judging when experiments were complete in a University of Washington lab. The key point is that this isn't a feature exclusive to one company's AI models, but a common specification that connects to any equipment with a programming interface, regardless of manufacturer or model.
See also
Stories using this term
- Anthropic unveils MHS, a standard for AI-controlled lab hardwareAI · 2026.08.28
- Anthropic launches Fable 5.1 and Mythos 5.1AI · 2026.09.02
- Anthropic releases second risk reportAI · 2026.08.15
- Reddit developer releases 'Unswarm' to auto-switch between multiple local LLMsAI · 2026.08.23
- Liquid AI unveils screen-reading model that runs in 3GBAI · 2026.08.13
- Open-source Ornith-1.5 claims scores on par with Claude OpusAI · 2026.08.21
