Anthropic's New Standard Lets AI Agents Run Lab Robots

Anthropic has opened a research preview of the Model Hardware Standard, a shared specification that lets AI agents operate physical laboratory and factory equipment through a single common interface. The standard, previewed on 27 August, aims to do for microscopes and robotic arms what earlier protocols did for software tools: give an agent one consistent way to discover and control a machine.

Most instruments in a lab do not talk to each other. Connecting them usually means weeks or months of bespoke integration work by specialists, as SiliconANGLE reported. The Model Hardware Standard, or MHS, is meant to cut that setup from weeks to hours by defining a common driver interface that many devices can speak.

Early results

The work began as a collaboration between Anthropic and the Howard Hughes Medical Institute's Janelia Research Campus. In one early test, researchers at Carnegie Mellon University used MHS to run a serial dilution experiment about three times faster than before, with a single agent orchestrating a liquid handler, a plate reader, a robotic arm and monitoring cameras spread across three computers that could not previously communicate, according to MarkTechPost.

Anthropic is releasing the preview to a first group of scientific labs and advanced manufacturers, and says it plans to publish MHS under an open-source licence. The framing is cautious. The specification puts weight on letting an agent operate devices safely, with the standard mediating what a model is allowed to do to hardware that can spill chemicals or swing a mechanical arm.

Why it matters

Software agents have spent the past year learning to browse, write code and file tickets. Reaching into the physical world is a harder problem, and a riskier one. If a common standard takes hold, the same agent that reads a paper could, in principle, set up and run the experiment it describes. That prospect sits alongside Anthropic's other science work, including its recent use of Claude to design working protein binders. It also revives a familiar question about oversight when the machine doing the pipetting answers to a model rather than a person.