Anthropic Lets AI Agents Run Lab Robots With a New Hardware Standard

Anthropic opened a research preview of the Model Hardware Standard on Thursday, August 27, a shared spec, built with HHMI Janelia Research Campus, so AI agents can safely operate physical devices, according to Anthropic.

Agents can run microscopes, liquid handlers, and robotic arms in parallel, covering drug discovery to laser calibration on a quantum computer.

It typically takes a lab weeks or months to wire hardware together because most devices do not talk to each other. MHS cuts that integration to hours or minutes with a standard driver and a reference file that tells an agent what the machine can measure, what it can change, and which safety limits it has to respect.

The spec is model-agnostic and works through MCP, a command line, or code and APIs, so any agent harness can use it, not just Claude, which is a different flavour of AI rollout than the video and travel tools Google has been stacking this week, from Gemini Omni 4K scene work to AI Mode flight tracking, and it sits closer to the bench than a notebook that can now pull in books you already own.

Genentech automated a BCA protein assay across a liquid handler, robotic arm, and plate reader. Carnegie Mellon ran dose-response experiments about three times faster, and QuEra’s agent recovered a quantum laser lock 99.3 per cent of the time without a human stepping in. University of Washington and Tetsuwan Scientific used it on qPCR work, the kind of bench job Canadian university labs still stitch together by hand.

Vendors adding support include AWS Strands Robots, Automata, Doosan, MBF Bioscience, QIAGEN, Tecan, and Universal Robots, with Danaher exploring a tie-in. Hugging Face’s LeRobot library and Raspberry Pi are early adopters, and Anthropic is taking a waitlist before it open-sources MHS with safety evaluations.