Anthropic Opens a Research Preview of the Model Hardware Standard (MHS): A Shared Specification for AI Agents to Safely Operate Physical Devices

for AI Agents to Safely Operate Physical Devices">

Anthropic has opened a research preview of the Model Hardware Standard (MHS) , a shared specification that lets AI agents discover and operate physical devices. The problem it targets is plumbing. A lab bench or factory cell is assembled from vendors that never planned to interoperate, so every instrument ships its own interface and specialists hand-write bespoke translators between each pair. According to Anthropic team, the setup normally takes weeks to months, and that MHS cuts it to hours or minutes.

Each instrument ships its own programming interface, so specialists hand-write bespoke translators between every pair. Even once wired together, there is no common way for devices to hand state to an agent or be operated safely by one. Anthropic says setup typically runs weeks to months; MHS reduces it to hours or minutes.

MHS standardizes the driver — the layer between an OS and a device. It exposes a small primitive set: read (get temperature), write (set temperature), plus discovery, so devices and agents find each other across a network without a translator in between.

It also carries knowledge code alone does not encode — the weight of a robot arm, for instance. Driver tags let a user write that in natural language, or have an agent interview them about the setup. The driver compiles those tags into a reference file: what a device measures, what can be adjusted, which safety limits are enforced.

Control runs through three mechanisms: the Model Context Protocol , a CLI, and code files. MHS is model-agnostic — any agent harness can reach it via standard protocols.

Genentech automated the BCA protein assay across a liquid handler, robotic arm, and plate reader. Claude ran trial transfers of dyed liquid, read absorbance, scored itself against an expert’s plate using RMSE, and converged on ~140 µL/s for water (0.016 RMSE) and 10 µL/s for viscous BSA (0.181 RMSE) — parameters its automation experts confirmed as reasonable.

QuEra Computing is the sharpest number. A bespoke laser-relock script, built over months by a four-person team, worked about 58% of the time at ~150 seconds per attempt. Handed the same problem through MHS, a four-role agent loop ran unattended overnight and produced a deterministic Python script that recovered the lock 695 times out of 700 — 99.3%, hardest cases in 10–14 seconds against 5–10 minutes for a human. Claude also cut the servo’s residual error from a specialist’s 15.7 mV to 1.55 mV; over a 19-hour run its tune never lost lock, while the expert tune unlocked about 1.6 times an hour ( QuEra blog ).