Image: Anthropic / anthropic.com Anthropic Opens a Research Preview for AI Agents to Run Lab Robots and Scientific Equipment
Anthropic's Model Hardware Standard (MHS) lets Claude and other AI agents safely operate physical lab devices — microscopes, liquid handlers, robotic arms — cutting integration time from months to hours.
Anthropic opened a research preview of its Model Hardware Standard (MHS) on August 27, giving developers in scientific research, robotics, and manufacturing a shared specification for connecting AI agents to physical devices.
The standard is model-agnostic. Any AI agent that speaks Model Context Protocol, command-line interfaces, or standard APIs can use it — not just Claude.
What MHS Actually Does
Most lab equipment doesn’t talk to other lab equipment. Getting a liquid handler, a robotic arm, and a plate reader to cooperate on a single experiment has historically required custom code and specialist engineers, with integration timelines measured in weeks or months.
MHS provides a common driver and reference file that communicates device capabilities, adjustable parameters, and safety constraints to an AI agent. The result: integration that used to take months now takes “hours or minutes,” according to Anthropic.
The spec was developed in collaboration with HHMI Janelia Research Campus and supports a broad range of programmable hardware — complex microscopes, liquid handlers, lasers, robotic arms, plate readers, qPCR instruments, and quantum computer laser calibration equipment.
Early Results
Early adopters have already demonstrated real speedups:
- Genentech automated a protein assay procedure coordinated across a liquid handler, a robotic arm, and a plate reader
- Carnegie Mellon University ran dose-response drug discovery experiments roughly three times faster than before
- QuEra Computing achieved 99.3% autonomous quantum laser lock recovery using an AI agent
Hardware vendors integrating MHS support include AWS Strands Robots, Automata, Doosan, MBF Bioscience, QIAGEN, Tecan, and Universal Robots. Hugging Face’s LeRobot library and Raspberry Pi are listed as early adopters, with Danaher exploring integration.
Why This Matters Beyond Labs
MHS marks Anthropic’s first serious push into physical AI — AI systems that take action in the real world through hardware rather than through software interfaces alone.
The research preview is intentionally scoped. Anthropic is releasing MHS to a first group of partners “to collaborate on building safety evaluations and developing best practices for AI systems operating physical equipment,” before making the standard open source. The emphasis on safety constraints built directly into the spec reflects how seriously the company is treating the transition from text generation to physical control.
For scientific organizations, the immediate value is throughput. Running experiments that previously required a human operator at each instrument step becomes possible autonomously, at any hour, across multiple parallel workflows.
The standard is available for review and early testing through Anthropic’s MHS research preview. The company plans to open source it after refining safety evaluations with initial partners.