Anthropic unveils standard letting AI agents control lab and factory equipment

Anthropic opened a research preview on August 27 of the Model Hardware Standard, a framework that gives AI agents a common interface to operate physical laboratory and manufacturing equipment such as microscopes, robotic arms, and liquid handlers. The standard, developed with HHMI Janelia Research Campus, aims to replace weeks of custom integration work with a setup process the company says can take hours or minutes.europeanbusinessreview+1

How It Works

MHS introduces a standardized driver layer between AI agents and programmable devices. Each connected machine gets a description file — written partly in natural language — that tells the agent what the device can measure, what parameters can be adjusted, and what safety limits apply. Agents interact with hardware through Anthropic's existing Model Context Protocol, a command-line interface, or code APIs, allowing them to sequence actions across multiple instruments simultaneously.fortuneindia

Anthropic compared the approach to USB-C: a single connector standard that lets different devices exchange information without bespoke wiring. The standard is model-agnostic, meaning it is not limited to Anthropic's Claude models.europeanbusinessreview+1

Early Results

Several organizations tested MHS before the public announcement. At Genentech, researchers connected a liquid handler, robotic arm, and microplate reader for a protein assay. Claude ran the experiment and adjusted parameters, finding that water worked best at roughly 140 microlitres per second while a more viscous protein solution required about 10 microlitres per second. At Carnegie Mellon University, a team built the required drivers in about eight hours — compared with several weeks for a typical vendor setup — and ran experiments roughly three times faster.fortuneindia

At quantum-computing company QuEra, an AI agent helped recover a laser's frequency lock. In a blind test involving 700 induced disturbances, the resulting script restored the lock 695 times, a 99.3 percent success rate.theneurondaily+1

What Comes Next

The research preview is available to a limited group of scientific research labs and advanced manufacturers. Early adopters include Hugging Face, which is adding MHS support to its LeRobot robotics library, and Raspberry Pi, which is enabling integration across some of its products. Anthropic plans to open-source MHS after using the preview phase to develop safety evaluations and determine where human approval should be required for higher-risk decisions.thedailystar+1

Anthropic acknowledged limitations. During the Genentech test, when bubbles caused errors, Claude initially repeated the operation with different parameters, making the problem worse. Researchers had to explain the physical cause before the model could respond appropriately. Elizabeth Kelly, Anthropic's head of beneficial deployments, told CNBC Comcast Corporation ↘ 0.95% that the launch is "a proof point rather than a finished commercial product," adding: "We built this for science to sort of show the promise of AI, but there's also huge benefits here for enterprise and for industry".fortuneindia+1