Anthropic gives Claude a standard for lab and robot hardware

Anthropic opened a research preview of the Model Hardware Standard, a shared specification for letting AI agents safely operate programmable physical devices. The company says the standard can connect instruments such as microscopes, liquid handlers and robotic arms, then expose them to agents through drivers, MCP, command-line tools and APIs.

The practical point is integration speed. Anthropic says labs and manufacturing teams often spend weeks or months wiring together devices with separate interfaces. MHS is designed to reduce that work to hours or minutes by making hardware discoverable and giving agents structured information about what a device can measure, adjust and safely do.

Anthropic says Claude has already used MHS-style setups to supervise experiments, update parameters in real time and recover from some hardware errors. Early examples include Genentech testing a protein-assay workflow, Carnegie Mellon running dose-response experiments about three times faster, and QuEra using an agent-built controller to recover quantum-laser lock 99.3% of the time.

The preview is also explicitly about safety. Anthropic says Claude still needs expert oversight for physical reasoning, and the research preview will be used to build evaluations and deployment guidance before MHS is open-sourced. Launch and early-access partners include AWS, Hugging Face, Raspberry Pi, Universal Robots, Tecan, QIAGEN and others.

Sources:

• https://www.anthropic.com/news/model-hardware-standard-research-preview