Anthropic opens a research preview of the Model Hardware Standard
A shared driver spec for agents to operate programmable lab and factory devices, starting with a closed partner group.
COMPANY // ACTIVE
ANTHROPIC.
AI lab behind Claude, the Model Context Protocol, and the Model Hardware Standard research preview.
NOTES
Primary source for MHS. Distinguishing company claims from independently verified lab results is part of every Anthropic-linked briefing.
SIGNALS
Independent write-up: MHS as a common driver layer for physical agents
Ars Technica describes MHS as standardized drivers so agents can talk to arbitrary devices without a custom translator for each instrument.
CNBC: Anthropic frames MHS as USB-C for machines, with an industrial follow-on
Business reporting adds competitive context — OpenAI and Amazon hardware spend, Anthropic's silicon hiring — around a still-closed research preview.
Hugging Face is adding MHS support in LeRobot
Anthropic names LeRobot as an early adopter for the next phase of device coverage.
Raspberry Pi: camera MHS driver tests, broader product integration planned
Another named hardware adopter in the Anthropic preview post, after successful Camera MHS Driver tests.
QuEra: agent-written laser lock recovers 99.3% of the time in MHS tests
A partner result from the preview post — specific, quantitative, and still vendor-mediated.
BRIEFINGS
Model Hardware Standard: agents get a driver layer for the physical lab
Anthropic's MHS research preview is not a robot and not a new model. It is a shared interface so agents can discover, describe, and operate programmable devices — with the hard problems still sitting in safety, spatial reasoning, and who writes the drivers.